OpenAI is projected to lose $5 billion this year.
In this podcast, Motley Fool analyst Jim Gillies and host Mary Long discuss:
Then Motley Fool analyst Asit Shama and host Ricky Mulvey test out a new rating system on Costco stock.
Like what you hear? You can vote for Motley Fool Money as Signal’s best money and finance podcast.
To catch full episodes of all The Motley Fool’s free podcasts, check out our podcast center. To get started investing, check out our beginner’s guide to investing in stocks. A full transcript follows the video.
This video was recorded on Oct. 03, 2024.
Mary Long: OpenAI sees some open check folks. You’re listening to Motley Fool Money. I’m Mary Long joined today by the illustrious Jim Gillies. Jim, thanks for being here. You ever been called that before?
Jim Gillies: That is absolutely the first time I’ve ever been called Illustrious, and I’m not sure what to do with it.
Mary Long: Thanks for joining us on the show. You are a valuation guy. I think it’s fair to say so figured we’d open up with some valuation news, OpenAI, which listeners might know that name. It’s the company behind ChatGPT, among other things. They got a new valuation, and that valuation comes out to 157 billion big ones. That is comparable to the likes of Goldman Sachs, Uber, AT&T. It’s also up from the $30,000,000,000 valuation it had about a year ago. I’ll open the floodgates. What do you think of that number?
Jim Gillies: Let’s say, why are you doing this to me? I think the number is absurd. I think the number is unjustifiable. I didn’t say wrong. I may very well be wrong. Maybe right, but the thing is I am from and the reason why I’m a little amused out of the gate is I am one of I think a fairly slowly dying breed. That is, I believe the company is worth the sum of its future cash flows, discounted back at an appropriate discount rate. Very important. Here’s the thing. I have no idea what ChatGPT OpenAI’s future cash flows will be. Neither does anyone else. I have no idea the timing of them, the scale of the size of them, the dilution that’s going to come. Neither does anyone else.
I have no idea what the appropriate discount rate is here. We can make an estimate, but it’s going to be precisely wrong, might even be roughly wrong, might be roughly right. This is clearly the big story of the world really, at this point, from a technology development standpoint. I don’t want to go in politics or anything. But AI is it’s transformational. We’re just not sure how transformational it will be. My history as an investor. We were talking before the show. We were talking a little bit about David Gardner and David Gardner’s style, which and I’ve many times said David Gardner is one of the best investors I’ve ever met. I’ve met Warren Buffett several times. But David’s style is something that often I struggle with because I’m not a venture capitalist thinker, whereas I think he is.
I would assume David would say Jim’s a cash flow thinker. I’m not a cash flow thinker, so I am seeing the world. David has that ability to see where the world is going at least better than I can and so I think it’s important to investor know thyself and follow along with what you are comfortable doing. All of that is a nice way to circle around and go like I would have, if this was public, I know they’re talking about going public at some point. I have no idea how you are applying this magical number to the valuation. We know that there’s no cash generation going on here. We have no idea where the cash is going to come, when it’s going to come. As Buffett likes to say, this goes into the two hard pile for me. It doesn’t go into the two hard pile for the Rule Breaker types, David Gardner types, and that’s great. But I would just walk away from this one because it doesn’t mesh with my style as an investor.
Mary Long: Another big news story circulating in business circles this week is about the port strike that’s happening. We had an e mail from a listener hit our inboxes earlier this morning. This listener, Andy did not know that I’d be chatting with you, but as luck would have it, he specifically asked for you in this e mail. Andy writes in, Andy here, longtime listener and member of both Motley Fool Stock Advisor and Rule Breakers, and also recently subbed to Motley Fool Canadian Services. I’ve got a question about the port strike that’s closed down ports on the US East Coast, and along the Gulf.
Specifically, I’m thinking of companies headquartered in Europe, which make auto parts like Autoliv ticker, ALV or Garrett Motion, ticker GTX. The Illustrious Jim Gillies has spoken about these companies over the past year and so I wonder if you might ask him, how much does he think the port strike will impact those companies at least in the short term and also whether the strike may create a buying opportunity.
Jim Gillies: Well, I don’t like thinking short term. But and Andy thank you for subscribing to the Canadian Services, which is why I’m answering the question. So, Autoliv is in 25 countries, including Canada. There’s a plant down near Windsor Ontario, which is where a lot of auto manufacturing is down near Windsor, Detroit crosses that border. They have worldwide facilities. The same goes for Garrett Motion. They’ve got worldwide facilities. Garrett’s got manufacturing ops in both California and Asia. Anything coming in from Asia is going to come in via the West Coast, not the East Cast. As well, automakers are notorious for their just in time manufacturing techniques.
Most likely, if a car is being assembled in North Carolina or Michigan or Southern Ontario, a lot of the parts are going into it are probably not all, and I’m not going to specifically say. Oh, yeah, Autoliv makes everything in Windsor for things being assembled in Detroit, because I don’t know that. But I would imagine that there is significant just in time issues in the manufacturing setup, that will alleviate some of this. Now, that’s said. I don’t want to minimize the potential impact of the strike. I think the strike could ultimately be very damaging. There’s some footage floating around of the Union lead, which is interesting. He is I think we’ll just say confrontational and aggressive, and I’ll leave it at that.
Mary Long: It’s fair.
Jim Gillies: If you’re a union member, I suppose that’s what you want and your leader, but it doesn’t look good, but I would warn against letting headlines define your investment choices unless unless, as Andy said here, this might be a short term opportunity. Because I think what Andy is not saying, but what I will say is strikes end, like strikes end and so if this does give you opportunity, and what we call this, headline risk gives you opportunities quite often. I could talk about the wisdom of recommending an aircraft Lasso during a pandemic, for example, which I did when all planes were shut down, international travel was shut down. The thesis can essentially be summarized as pandemics end. I don’t know when, probably be a Tuesday, but they end, and when they end, we’ll probably still take planes. The world’s largest aircraft also that would be AerCAP, by the way. The world’s largest aircraft or so will probably have pretty robust demand when it happens.
Mary Long: Less front and center than OpenAI’s, new valuation, and the port strike. Constellation Brands reported this morning? This is the company behind Corona, Modelo, Svedka Vodka, Robert Mondavi Wine. Some of the top line news from their earnings report, overall sales rose. That’s mostly due to their beer business. They saw a lot of weakness in the Wine and Spirits business. They’re also writing off a $2.25 billion impairment loss for that Wine and Spirits business. First up, Jim, what’s eating Constellation’s wine business?
Jim Gillies: I feel that I’m here under false pretenses because I thought were going to talk about Constellation Software, which I hold is the best company in Canada. But no, it’s Constellation Brands.
Mary Long: Wamp.
Jim Gillies: I know, yeah, sad trombone for me.
Mary Long: Maybe be less exciting.
Jim Gillies: Yeah. Look, maybe it’s just a mature industry. Maybe it’s a mature business. I promise this is going to go somewhere. I’ve been to New Zealand. I’ve toured around New Zealand and all the wineries. the one I could not have cared less about, Kim Crawford Wine. Why? ‘Cause it’s big industrial wine made by you guest at Constellation Brands. I think, this is all just going to be riffing here, but maybe people beset with choice are choosing elsewhere. I can tell you, in this house, we don’t drink a lot of wine. Consumption and has been consumption has been going down, just because I think we find as we get older, we don’t appreciate it that much or I don’t want to drop an extra $20 on $20. Someone out there just said $20, Gillies. That’s a terrible bottle of wine, you get to spend up.
We probably drink half or less of what we drank a decade ago. To think I appreciate good wine. Someone professional might call what I like swill, but that’s fine. But, we go to wineries. we’ll go to Niagara, which is fairly close. We’ll go out to British Columbia when we’re there. We’ll go to wineries. I’ve been to wineries in Virginia I’ve been to wineries in California. Like and through the magic of the Internet, I can order what I like if it’s a winery I went to in Napa, and it was a smaller winery. It’s not a big giant corporate winery. But so maybe one thing is people just don’t like mass production wines anymore. Another thing and this might be fairly Canada focused, but I know it’s happening a lot more in your fine country is, you may have heard that cannabis was legalized in Canada across Canada in 2018. One of the things that’s happened as a result of that is there’s a lot of cannabis based beverages, like for people who I guess who don’t want to smoke or whatever, which strikes me as an intelligent choice. For people who are using wine just to relax and laugh with friends or just take the edge off, maybe they’re not drinking wine anymore.
Maybe things have migrated to cannabis beverage versus and wine has become a beverage that you drink with your sar duck breast or rare flank state thing. It’s become more of I don’t know anybody. I’ve not been to a lot of dinner parties since the pandemic, but I’d be pretty surprised if someone served me a cannabis beverage when I went to a dinner party. Then you’ve also got that look, we know with the data saying that younger people are drinking less. those just coming into coming out of their slothful teenage years and into their early 20s and legal drinking age. They’re drinking less and they’re opting away from wine. I can tell you. We had a thing in my house about two years ago with my son who has just turned 20, but drinking age here is 19. There was a medical family situation and I had to call my son. He was at University at the time, and I said, I might need to come get you. It was a Friday night. No drinking.
I have fun, but I might have to come get you because something could go wrong tonight and so I would appreciate you and my son, first off he said, well, Dad, I’m actually at my friend’s house, and the friend that he has is muslims there’s no alcohol in the house, anyway. Then my son is also a body builder. Then I got a five minute pseudo lecture on how alcohol consumption is really detrimental for body builders. I’m like, dude, dude. I appreciate all that. Family medical emergency. I don’t need that now. Stand by and hopefully, I don’t have to come get you. But it’s like, kids, and I see his friends. I see my daughter’s friends, and it’s like I think they’re being more responsible than I was. How about I leave it at that?
Mary Long: So we talk about shifting consumer taste. We mentioned cannabis beverages. This is a consumer brands business. Again, consumer taste change over time. Is the diversification of constellations portfolio, which is a big portfolio? Is that diversification a strength or a weakness here? Because the more things you offer, chances are that some are not going to perform so well.
Jim Gillies: Well, yeah, and the more things you offer, you dilute you want to find the one thing you’re really great at or the two or three things you’re really great at. I often find that when you offer so many choices, none of them will really move the needle, and some will be a disappointment this quarter, and some will be a disappointment next quarter. The other thing is, two, and I was an investor for a brief period of time in Moles encores. Spoiler didn’t go terribly well.
But competitor, and we know that Anheuser-Busch InBev ASB Miller, whatever the hell, they’re calling themselves nowadays. This tends to be a scale business and there’s all these companies, Constellation, Anheuser-Busch, Molson Coors. They are just constantly shoving money at at new CapEx and acquisitions, and Maybe the industry is just largely topped out because none of them have been good strong investments. Again, did I mention my Molson Coors foray did not go terribly well. They are their old brands, but, like, there’s no one’s going to challenge beverauge on scale. Maybe Heineken can go and buy Constellation and Acres, but that might bring them within spitting distance, but then it’s not a high growth industry and so it doesn’t really matter to me how many brands you’ve got.
I think most of these companies where they did try to maybe make a foray into the cannabis area, I think most of them lost their hats on that one. Maybe that’s where you should be going, but then again I would say. Well, it’s called weed for a reason, and it’s a commodity product. I think the best thing to do in this space is to say, you know what? I’m going to avoid individual stock risk in this area. I think what I’m going to do is if there’s some exposure in an S&P 500 index fund, I’m going to say, you know what? I’ve got exposure, and I’m happy with that level of exposure.
Mary Long: Jim Gillies, thanks for joining us today on Motley Fool Money. Always a pleasure to have you even though I feel like we were pretty bearish throughout the entirety of today.
Jim Gillies: I don’t think so. I don’t want to be bearish. I think Autoliv and Garrett will be fine. How’s that? I own both.
Mary Long: I set you up for that a little bit because we do have Ricky and Asit talking about Costco in the segment after us and that is something you are certainly not bearish on and I didn’t even give you the opportunity.
Jim Gillies: That is very correct, yes.
Mary Long: I apologize to you.
Jim Gillies: Perfectly fine, and I don’t own enough Costco.
Mary Long: Before we get to the next segment, a humble request. Motley Fool Money is a finalist for Signal’s Best Money and finance broadcast. We’re up against some really great shows from Barns, the Financial Times, and Bloomberg. The winner, though, is determined by your vote. So, if you enjoy this show, all of us here at Motley Fool Money would really appreciate you taking a moment to cast your vote for us. I’ll drop a link in today’s show note so that you can do that if you’re so inclined. Heads up, you’ll need to enter your email to verify that you’re a real person, not a robot. Thanks for helping us out.
As always, we appreciate you listening. Already up next, Costco gets a lot of love, not just from Jim Gillies, but from many investors, but it also comes at a pretty price. Asit Sharma and Ricky Mulvey test out a new rating system on Costco’s valuation.
Ricky Mulvey: We’ve heard the booms. We’ve heard the dooms around the Costco Food Court, the chicken bake, the double chunk chocolate cookie, Asit. We’re not food reviewers, though, so we’re going to look at the stock. But before we do so, the most important question of this segment. Are you a Costco guy? Do you get it?
Asit Sharma: Ricky, what’s a Costco guy? I guess I get it. I used to shop at Costco when our kids were younger, and the bulk purchases made a lot of sense for us.
Ricky Mulvey: We have sort of the duality of Costco right now because the food court is so inexpensive. Hot dog combo, dollar 50. We walk out of Costco, thinking about the great deals we got, three pounds of coffee for, like, 15 bucks. When we look at the stock though it’s a different story. It’s at one of the most expensive multiples it’s ever been. Trades around 55 times free cash flow, 31 times enterprise value to forward EBITDA so the value of its debt and equity compared to its forward earnings. Both of these are high historically. Then when you compare it to its peers, Kroger, for example, is less than half of that valuation for its free cash flow, and as opposed to the 31X earnings multiple, it’s at a 6X. Why does Costco get such a premium as opposed to these other grocery stores?
Asit Sharma: Ricky listening to you, I was just thinking, it doesn’t matter what batter you dip this thing into, you put it in the oven. It’s double chunk all the way right now. It’s just going to be expensive under any valuation measure when you pull that out of the analysis oven. This is a little mysterious because if you compare Costco to traditional large scale retailers grocers, let’s say, it looks very expensive. It should be. It’s got the membership that has renewal rates above 90%. They’ve gone on this big cost optimization drive, which has really helped them increase margins. They’re the beneficiary of lower input costs going forward as inflation eases. They have shifted some business to high ticket items, big appliances, which carry a better margin and just bigger absolute dollars for the company as they’ve built out more logistics and distribution. They’ve got the balance sheet for global expansion.
Asit Sharma: When you put all this together, you start thinking, yeah, I mean, it should be more expensive than some of these traditional grocers. But then look at a company like BJ’s Wholesale Club, which has, not all of those advantages, but has been undergoing some of the same transitions and has mirrored the stock price over the last five years, BJ’s trades at a fraction of what Costco trades at under the same multiples. I think the reason is the stability of the cash flows is something that the market is looking at as a forever proposition, and they love the way the company allocates its capital between all this expansion and then special dividends that come up every few years and its ability to keep those loyal members and raise prices on them every few years. Still, it just seems historically expensive right now. The market is looking past a lot of potential near term issues to say, we’re going to hold the stock. We’re actually going to act like foolish investors maybe. Still, I would say, if you don’t own Costco, you could nibble. It’s just pricey here.
Ricky Mulvey: What are the near term issues that you’re thinking about?
Asit Sharma: Well, number 1 is the volatility of those same input prices. The market expects that deflation is at least a deceleration of inflation, which the Fed is seeing also is going to benefit companies like Costco which have become more efficient as prices are rising. On the back end, when things get a little cheaper, the inference is that these companies will keep some of that margin to themselves. But look, we have potentially a port strike coming up, so you can’t hang your hat some near term tailwinds. That’s just one of them. I think the other is that we don’t know yet if we’ll really come out of this current economic environment with a soft landing. It always feels like it. We had a shift in interest rate posture by the Fed. Inflation has been easing a bit, but that’s not to say that we won’t fall into a mild recession.
A mild recession might make people think a little bit differently about that 55X ford multiple on earnings per share. But at the end of the day, I think so many institutional buyers and retail buyers are saying, this is a quality company. We don’t mind overpaying a little bit just now. Costco historically, actually, as you pointed out, maybe doesn’t trade this high, but it’s always expensive. There could be some relative biting the bullet here for people who want to be in on this company.
Ricky Mulvey: We’re talking more about round to at the Fool. Let us not do a full re litigation on what this measures. We did a segment on it a while back. But basically, it’s a measure of how efficient a company is a generating income on the hard assets that it has. When you look at Costco, it’s got about 11 billion in cash. It’s got a lot of land in stores. It’s got about 17 billion in inventory, all of those warehouse goods moving through its supply chain on the way to the large boxes at the end of checkout. But when you look at Costco’s round to how efficient it is, it’s at 26%. That’s actually slightly below Kroger and Walmart.
With the premium we just talked about, and Costco investors are willing to pay for the stock, I’m surprised it’s not significantly higher than these other grocers which have more items that they’re selling that don’t have that wonderful membership loyalty program, which completely cuts out on things like shoplifting at their store, what say you? What’s going on here with Costco’s efficiency at generating income?
Asit Sharma: Well, I think that all three that you mentioned are pretty efficient for being primarily grocers. What it might be going on here is a bit of investment. When you invest in your capital base and make it bigger, that lowers your return on invested capital. If you’re a company like these grocers, which doesn’t have a lot of tangible assets on your books, your round to is anyway going to be closer to your return on invested capital. Actually, round to might not be the most efficient metric to use here. But, since not any of these three companies really has, let’s say, a huge goodwill component on their books or a lot of amortization for intangibles, let’s compare these apples to apples.
What is going on here? Is that expansion of the base of warehouses with Costco? Take Kroger, for example, Kroger has 60% of Costco sales, but they have almost tripled the amount of leased assets on their books. What that means is that Costco is buying land building stores, that’s a bigger base, so the bigger the base, the less return penny for penny on your income that you bring home. That’s actually what you want to see if you are an investor in Costco. You want to say, look, take that cash, buy some more land, buy some more buildings, build some more buildings. If you look at their latest supplemental presentation for this previous quarter just end. They have a nice picture of their Nanjing China warehouse. It’s humongous, and they have one that’s in Chungsan North Korea, I hope I pronounced that correctly. Also humongous. As a shareholder, you want that asset base to be big and get bigger because this is a company that’s got to scale up. It’s so big and so mature, Ricky. The only way to keep delivering those returns, aside from the cost optimization and new SKUs, new things that you and I can buy is that global expansion.
Ricky Mulvey: So the thing that’s pressuring it a little bit is that it owns a lot of its real estate that it has a lot of assets. You know Asit, maybe one of these days, we can do retailers with surprising routes. Hey, how about Dillards at 42%? That’s a separate topic.
Asit Sharma: That is surprising.
Ricky Mulvey: As we focus on Costco’s valuation one thing that the Costco guys, beause really what they’ve done has provided gifts for us is social media users. They fundamentally changed my life philosophy. I now sort stocks, my relationships, TV shows, things I watch, either into boom, five booms, that’s boom, boom, boom, boom, boom, or Doom. Boom, five booms or doom. Where are you at on Costco’s valuation as a stock?
Asit Sharma: Boom. Boom, in the sense of that is surprising, but not worth five booms, which is I’ll keep buying it. I don’t care what the price is. You have to be rational a bit.
Ricky Mulvey: Then as we look at Costco’s recent decisions, I’m going to let you be the manager, the CEO. You’re giving feedback, and the only way you’re giving feedback is boom, five booms or doom. There’s three of them. One, they’ve recently raised their annual membership fees by 5-$10. Boom, five booms or doom.
Asit Sharma: How about a one boom?
Ricky Mulvey: It’s a sad boom you offered. You were like, one boom.
Asit Sharma: A one boom it’s not a doom. Costco raises its membership fees only every 6-7 years. I actually expected a little more than these marginal improvements, but it speaks to the power of the model. Actually, this was applauded in a lot of circles because they’re doing right by members and not jacking up the price so much that people want to shop at other places. Throw the next one at me.
Ricky Mulvey: You mentioned the high ticket items earlier. Gold bullion. Now Costco is a dealer in gold bullion. If you look at the jewelry cases the next time you’re at Costco, you’re going to find these one ounce bars of gold that are being sold at cost. Boom, five booms or doom.
Asit Sharma: Boom, boom, boom, boom, boom Ricky.
Ricky Mulvey: Full five.
Asit Sharma: Five booms on this idea. I love this idea. The reason I love it so much is that it is so unexpected if you don’t follow Costco. What is Costco doing selling gold bullion? Their membership really took that offer in, and you see what Costco is doing here is perhaps sowing the seeds of something they can do over and over again with surprising items. When you can harness the purchasing power, outside of membership, outside of regular visits to the store, outside of e-commerce, to stuff that people almost virally, your members want to have. That is very powerful to your business model. It may seem silly, but it’s signaling something that is, I think quite valuable to Costco in years to come. I was surprised by that. It’s a five boom to me.
Ricky Mulvey: As we wrap up, let’s take a look at the latest Costco earnings call. We got some news from them. Comp sales up 6%. That’s pretty good for a grocery store. We’re also seeing that pricing power is. Costco boneless chicken tenders. They actually lowered the price by 13% and saw a 21% lift in the volumes sold. Also, Costco’s app getting bigger with 3.5 million app downloads just in the quarter. Anything meaningful here for the business of Costco. We’ve talked about the valuation. Anything meaningful for the business is Costco keeps chugging along.
Asit Sharma: I think all three are relevant. Comp sales, that combination of good volume traffic trends, being able to have the merchandise people want. That’s so important. The more mature company gets, typically, the more those comp sales will trend toward 1-2% or around the cadence of long-term inflation, which forget about recent inflation, has been 2-3% over time. So, for company this big, whenever you can have comp sales above the 5% level, you’re doing really well. The idea that Costco’s buyers and the people who shape its inventory are like those at TDMax, which are the best and brainiest in the business is something I think that’s not as appreciated out in the investing world. Being able to have the inventory on hand, know what people are really going to buy and then drop the price and sell in volume is pretty nice. I think those Apple Out downloads are also meaningful for Costco, although it’s so big, it’s not like a immediate needle mover, but still it’s quite good. That’s a very, very decent number.
Ricky Mulvey: So, I will count you as impressed with the business, but maybe a little concerned about the valuation. I will wrap this up with an om. One half of a boom as we split the difference. Asit Sharma thanks for being here. Appreciate your time and your insight.
Asit Sharma: Thanks a lot, Ricky. Always fun to be here.
Mary Long: As always, people on the program may have interest in the stocks they talk about, and the Motley Fool may have formal recommendations for or against, so don’t buy or sell stocks based solely on what you hear. I’m Mary Long. Thanks for listening. We’ll see you tomorrow.
¿Operai está desarrollando una plataforma de redes sociales?
Según los informes, Openai está desarrollando una plataforma de redes sociales similar a X, con un prototipo interno que integra las capacidades de generación de imágenes de ChatGPT.
queso Cheddar
SAN FRANCISCO – El concurso en Silicon Valley para dominar la inteligencia artificial se está desarrollando en una nueva corte: los investigadores superestrella.
Si bien la lucha para atraer el mejor talento y mantenerlos felices siempre ha sido un sello distintivo de la industria tecnológica, desde que ChatGPT se lanzó a fines de 2022, el reclutamiento se ha intensificado a los niveles profesionales de atletas, una docena de personas que han estado involucradas en el reclutamiento de investigadores de IA dijeron a Reuters.
“Los laboratorios de IA se acercan a la contratación como un juego de ajedrez”, dijo Ariel Herbert-Voss, CEO de la startup de ciberseguridad Runsybil y un ex investigador de Operai que ingresó a la pelea de talentos después de lanzar su propia compañía. “Quieren moverse lo más rápido posible, por lo que están dispuestos a pagar mucho por candidatos con experiencia especializada y complementaria, al igual que las piezas del juego. Son como, ‘¿Tengo suficientes torres?
Empresas, incluidas OpenAi y Google, ansiosas por obtener o mantenerse por delante en la carrera para crear los mejores modelos de IA, cortan a estos llamados “IC”: los contribuyentes individuales cuyo trabajo puede hacer o romper empresas.
Noam Brown, uno de los investigadores detrás de los recientes avances de IA de OpenAi en el razonamiento complejo de matemáticas y ciencias, dijo que cuando exploró las oportunidades de trabajo en 2023, se encontró siendo cortejado por la élite de Tech: el almuerzo con el fundador de Google Sergey Brin, póker de Sam Altman’s y una visita de avión privado de un ansioso inversor. Elon Musk también hará llamadas para cerrar candidatos para Xai, su compañía de IA, dijeron que dos personas que han hablado con él. Xai no respondió a una solicitud de comentarios.
Finalmente, dijo Brown, eligió OpenAi porque OpenAi estaba dispuesto a poner recursos, tanto las personas como el cálculo, detrás del trabajo que estaba entusiasmado.
“En realidad, no era financieramente la mejor opción que tenía”, dijo, explicando que la compensación no es lo más importante para muchos investigadores. Eso no ha impedido que las empresas arrojen millones de dólares en bonos y paquetes de paquetes a los investigadores STAR, según siete fuentes familiarizadas con el asunto.
Algunos investigadores de los principales abiertos que han indicado interés en unirse a la nueva compañía de la ex científica jefe Ilya Sutskever, SSI, se les ofreció bonos de retención de $ 2 millones, además de aumentos de capital de $ 20 millones o más, si se quedaron, a dos fuentes le dijeron a Reuters. Algunos solo se les ha requerido que se queden durante un año para obtener la bonificación completa. SSI y OpenAi declinaron hacer comentarios.
Otros investigadores de Operai que han presentado ofertas de once laboratorios han recibido bonos de al menos $ 1 millón para quedarse en OpenAI, dijeron dos fuentes a Reuters. Los principales investigadores de OpenAI reciben regularmente paquetes de compensación de más de $ 10 millones al año, dijeron las fuentes.
Google Deepmind ha ofrecido a los principales investigadores $ 20 millones por año paquetes de compensación, otorgados con subvenciones de capital fuera del ciclo específicamente a los investigadores de IA, y también ha reducido la adjudicación de algunos paquetes de valores a 3 años, en lugar de los 4 años normales, dijeron las fuentes. Google declinó hacer comentarios.
Por el contrario, los principales ingenieros de Big Tech Companies reciben una compensación anual promedio de $ 281,000 en salario y $ 261,000 en capital, según CompreHensive.io, una compañía que rastrea la compensación de la industria tecnológica.
Si bien el talento siempre ha sido importante en Silicon Valley, la diferencia con el auge de la IA es cuán pocas personas están en este grupo de élite, dependiendo de a quién le pregunte, el número podría variar de unas pocas docenas a alrededor de mil, ocho fuentes dijeron a Reuters.
Eso se basa en la creencia de que este pequeño número de “IC” ha hecho contribuciones de gran tamaño al desarrollo de grandes modelos de idiomas, la tecnología en la que se basa el auge de IA de hoy y, por lo tanto, podría hacer o romper el éxito de un modelo de IA.
“Seguro que los ingenieros 10X son geniales, pero maldita sea esos 10,000x ingenieros/investigadores …”, tuiteó el CEO de OpenAI, Sam Altman, a finales de 2023, aludiendo a una máxima larga que los mejores ingenieros de software fueron 10 veces más buenos que el promedio (10X), pero ahora en la industria de la IA, los mejores investigadores son 10,000 veces (10,000x) tan efectivos que el promedio.
La partida de septiembre del director de tecnología de OpenAi, Mira Murati, quien luego fundó una startup rival de IA, ha intensificado la Guerra de Talento AI. Murati, conocido en Operai por sus habilidades de gestión y destreza de ejecución, reclutó a 20 empleados de Operai antes de anunciar su empresa en febrero. Ahora ha atraído aún más investigadores de Operai y otros laboratorios, y el equipo ahora tiene alrededor de 60 personas, dijeron dos fuentes a Reuters. Aunque la compañía no tiene ningún producto en el mercado, Murati está en medio de cerrar una ronda de semillas récord que se basa en la fuerza del equipo. Un representante de Murati declinó hacer comentarios.
La escasez de talento ha obligado a las empresas a acercarse a la contratación creativamente. Zeki Data, una empresa de datos centrada en identificar el talento de IA de los mejores IA, dijo que está empleando técnicas de análisis de datos de la industria del deporte como la popularizada por la película “Moneyball” para identificar talento prometedor pero no descubierto. Por ejemplo, los datos de Zeki descubrieron que Anthrope ha estado contratando investigadores con antecedentes de física teórica, y otras compañías de inteligencia artificial han contratado individuos con antecedentes de computación cuántica.
Anthrope no respondió a una solicitud de comentarios.
“En mi equipo, tengo matemáticos extraordinariamente talentosos que no habrían venido a este campo si no fuera por el rápido progreso que estamos viendo ahora”, dijo Sébastien Bubeck, quien dejó su papel como vicepresidente de investigación de Genai en Microsoft el año pasado para unirse a Openi. “Estamos viendo una afluencia de talento de todos los campos que están en IA ahora. Y algunas de estas personas son muy, muy inteligentes, y marcan la diferencia”.
Informes de Anna Tong en San Francisco; Edición de Kenneth Li y Claudia Parsons
Los chatbots de IA avanzan rápidamente y los probando hasta sus límites es lo que hago para vivir. El soneto Claude 4 de Anthrope y el chatgpt-4o de OpenAI son dos de las herramientas más inteligentes disponibles en este momento. Pero, ¿cómo se comparan realmente en el uso diario?
Para averiguarlo, le di a ambos modelos el mismo conjunto de 7 indicaciones; Cubriendo todo, desde narración de cuentos y productividad hasta apoyo emocional y pensamiento crítico.
“Escríbeme un resumen conciso de Misión imposible Personajes y tramas hasta la fecha ”, recientemente le pregunté a ChatGPT antes de atrapar la última entrada de franquicia. Se entregó. No necesitaba entender su código o conocer su conjunto de datos de capacitación. Todo lo que tenía que hacer era preguntar.
ChatGPT y otros chatbots impulsados por modelos de idiomas grandes, o LLM, son más populares que nunca. Los científicos están tomando nota. Las proteínas, los caballos de batalla moleculares de las células, mantienen nuestros cuerpos corriendo suavemente. También tienen un idioma propio. Los científicos asignan una letra abreviada a cada uno de los 20 aminoácidos que componen proteínas. Al igual que las palabras, las cadenas de estas letras se unen para formar proteínas de trabajo, su secuencia determina la forma y la función.
Inspirados en LLM, los científicos ahora están construyendo modelos de lenguaje de proteínas que diseñan proteínas desde cero. Algunos de estos algoritmos están disponibles públicamente, pero requieren habilidades técnicas. ¿Qué pasaría si su investigador promedio podría simplemente pedirle a una IA que diseñe una proteína con un solo mensaje?
El mes pasado, los investigadores dieron a Protein Design AI el tratamiento con chatgpt. De una descripción del tipo, estructura o funcionalidad de una proteína que está buscando, el algoritmo produce posibles candidatos. En un ejemplo, la IA, denominada pinal, hizo con éxito múltiples proteínas que podrían descomponer el alcohol cuando se analizó dentro de las células vivas. Puedes probarlo aquí.
Pinal es el último en un creciente conjunto de algoritmos que traducen el inglés cotidiano en nuevas proteínas. Estos diseñadores de proteínas entienden el lenguaje sencillo y la biología estructural, y actúan como guías para los científicos que exploran proteínas personalizadas, con poca experiencia técnica.
Es un “enfoque ambicioso y general”, el equipo internacional detrás de Pinal escribió en una preimpresión publicada en Biorxiv. La IA aprovecha el “poder descriptivo y la flexibilidad del lenguaje natural” para hacer que las proteínas de diseñador sean más accesibles para los biólogos.
Enfrentados contra los algoritmos de diseño de proteínas existentes, Pinal entendió mejor el objetivo principal de una proteína objetivo y aumentó las posibilidades de que funcionaría en las células vivas.
“Somos los primeros en diseñar una enzima funcional usando solo texto”, dijo Fajie Yuan, científica de IA de la Universidad de Westlake en China que dirigió el equipo. Naturaleza. “Es como la ciencia ficción”.
Las proteínas son los componentes básicos de la vida. Forman nuestros cuerpos, el metabolismo del combustible y son el objetivo de muchos medicamentos. Estas intrincadas moléculas comienzan a partir de una secuencia de “letras” de aminoácidos, que se unen entre sí y eventualmente se doblan en intrincadas estructuras 3D. Muchos elementos estructurales, un bucle aquí, un tejido o bolsillo allí, son esenciales para su función.
Los científicos han intentado durante mucho tiempo diseñar proteínas con nuevas habilidades, como enzimas que descomponen de manera eficiente los plásticos. Tradicionalmente, han personalizado las proteínas existentes para un cierto uso biológico, químico o médico. Estas estrategias “están limitadas por su dependencia de las plantillas de proteínas existentes y las limitaciones evolutivas naturales”, escribieron los autores. Los modelos de lenguaje de proteínas, en contraste, pueden soñar con un universo de nuevas proteínas sin ataduras de la evolución.
En lugar de absorber el texto, la imagen o los archivos de video, como LLMS, estos algoritmos aprenden el lenguaje de las proteínas entrenando en secuencias y estructuras de proteínas. ESM3 de EvolutionaryScale, por ejemplo, entrenó en más de 2.700 millones de secuencias de proteínas, estructuras y funciones. Los modelos similares ya se han utilizado para diseñar anticuerpos que luchan contra ataques virales y nuevas herramientas de edición de genes.
Pero estos algoritmos son difíciles de usar sin experiencia. Pinal, por el contrario, apunta al científico promedio-joe. Al igual que una cámara DSLR en Auto, el modelo “evita las especificaciones estructurales manuales”, escribió el equipo, lo que hace que sea más simple hacer su proteína deseable.
Para usar Pinal, un usuario le pide a la IA que construya una proteína con una solicitud de varias palabras clave, frases o un párrafo completo. En la parte delantera, la IA analiza los requisitos específicos en el aviso. En el back -end, transforma estas instrucciones en una proteína funcional.
Es un poco como pedirle a ChatGTP que le escriba una reseña de restaurante o un ensayo. Pero, por supuesto, las proteínas son más difíciles de diseñar. Aunque también están formados por “letras”, su forma final determina cómo (o si) funcionan. Un enfoque, denominado entrenamiento de extremo a extremo, traduce directamente un aviso en secuencias de proteínas. Pero esto abre la IA a un vasto mundo de secuencias potenciales, lo que hace que sea más difícil marcar las secuencias precisas de las proteínas de trabajo. En comparación con las secuencias, la estructura de proteínas, la forma 3D final, es más fácil para el algoritmo generar y descifrar.
Luego está el dolor de cabeza de los datos de entrenamiento. Aquí, el equipo recurrió a las bases de datos de proteínas existentes y usó LLM para etiquetarlas. El resultado final fue una vasta biblioteca de 1.700 millones de pares de texto proteico, en el que las estructuras de proteínas coinciden con descripciones de texto de lo que hacen.
El algoritmo completado utiliza 16 mil millones de parámetros, estas son las conexiones internas de una IA, para traducir el inglés simple al idioma de la biología.
Pinal sigue dos pasos. Primero traduce las indicaciones en información estructural. Este paso divide una proteína en elementos estructurales, o “fichas”, que son más fáciles de procesar. En el segundo paso, un modelo en idioma proteico llamado Saprot considera la intención del usuario y la funcionalidad de proteínas para diseñar secuencias de proteínas con mayor probabilidad de doblar en una proteína de trabajo que satisfaga las necesidades del usuario.
En comparación con los algoritmos de diseño de proteínas de última generación que también usan el texto como entrada, incluida ESM3, el pinal superó la precisión y la novedad, es decir, generar proteínas no conocidas por la naturaleza. Usando algunas palabras clave para diseñar una proteína, “la mitad de las proteínas de pinal exhiben funciones predecibles, solo alrededor del 10 por ciento de las proteínas generadas por ESM3 lo hacen”.
En una prueba, el equipo le dio a la IA un breve aviso: “Por favor, diseñe una proteína que sea una alcohol deshidrogenasa”. Estas enzimas descomponen el alcohol. De más de 1.600 proteínas candidatas, el equipo eligió los ocho más prometedores y las probó en células vivas. Dos rompieron con éxito el alcohol a temperatura corporal, mientras que otros fueron más activos a un sudor de 158 grados Fahrenheit.
Las indicaciones más elaboradas que incluían la función de una proteína y los ejemplos de moléculas similares, arrojaron candidatos a antibióticos y proteínas para ayudar a las células a recuperarse de la infección.
Pinal no es el único IA de texto a proteína. El Startup 310 AI ha desarrollado una IA denominada MP4 para generar proteínas a partir del texto, con los resultados que la compañía dice que podría beneficiar la enfermedad cardíaca.
El enfoque no es perfecto. Al igual que los LLM, que a menudo “alucinan”, los modelos de lenguaje de proteínas también sueñan secuencias poco confiables o repetitivas que reducen las posibilidades de un resultado final de trabajo. La redacción precisa de las indicaciones también afecta la estructura de proteína final. Aún así, la IA es como la primera versión de Dall-E: juega con ella y luego valida la proteína resultante usando otros métodos.
Remove.bg: La Revolución en la Edición de Imágenes que Debes Conocer
Cómo Comenzar a Utilizar ChatGPT: Una Guía Completa para Principiantes
Startups de IA en EE.UU. que han recaudado más de $100M en 2024
Cómo Empezar con Popai.pro: Tu Espacio Personal de IA – Guía Completa, Instalación, Versiones y Precios
Deepgram: Revolucionando el Reconocimiento de Voz con IA
Perplexity aplicado al Marketing Digital y Estrategias SEO
Curso de Inteligencia Artificial de UC Berkeley estratégico para negocios
Google Gemini: Guía Completa para Comenzar a usarlo y ganar tiempo en nuestras tareas diarias
Inside the OpenAI-DeepSeek Distillation Saga & Alibaba’s Most Powerful AI Model Qwen2.5-Max
De ChatGPT a mil millones de agentes
