At this point, arms-crossed mugshot of "{ex-Open AI} raises $2B" is a meme. Training one epoch of a model the size of the GPT-4 family is easily an 8 figure job. Considering other ops costs, UI and backend dev, securing Everest sized datasets, demos/talks, wages, bonuses etc, this probably gives them around 100 trial and errors, or in other words "Look at me mum, I won 100 chips for the casino!"
And imagine being the ops guy there, about to run a new training batch, but you specified the wrong input path. Puff, the money is gone. Absolutely wild stuff.
I think it takes a lot of temerity and hubris to look at a 2B raise and assume it's all hype. A16Z has certainly had some misses, but one assumes there actually is some product they're showing behind closed doors that makes this round much more reasonable than it appears from the outside.
If something makes no sense, seems totally crazy, and is being done by a crowd of extremely smart people, you can only assume one of two things: they are actually crazy and frittering away 2B on hype or, just maybe, there's something we're not aware of. If there are only two camps: optimistic and naive or pessimistic and dismissive, I'll choose naive every day of the week.
Anyways, congrats to Thinking Machines and here's hoping they do have something awesome up their sleeve!
> they are actually crazy and frittering away 2B on hype
This assumes that being careless with billions can only ever be crazy.
If you're already set for life, why not gamble (including at completely irrational levels) for even more insane amounts of money when the whole thing is just a crazy house of mirrors. Yes, there's value at the heart, but there's also crazy amounts of money being funneled in, lots of opportunity for chaos, lots of chances for legal rug pulls. All of it inflated even further by a fervor of carelessness for any kind of consequences - things like the stock market are completely removed from any kind of fundamentals.
In a fun house of mirrors, that 2 billion could be 2 cents, or it could be 2 trillion. Buy the ticket and have fun!
I don't disagree with your broader point, but have spent enough time with enough a16z partners to say they're just people. Not outright stupid, but not extremely smart, either. And their error rate is pretty high.
Which...to some extent is by design. It's part of a VC's job to make bad bets. Sometimes the price of getting into a deal at all is getting in on insane terms, but you still do it because that one investment could return the entire fund. Maybe Thinking Machines is a winner, maybe it's another Clubhouse. We'll see.
I do agree for $2bn, but it’s also well known that US and in particular SV VCs will fund ideas and people, I.e. potential, whereas everywhere else funds results.
It’s that culture that creates some spectacular hits, and a vast number of misses. Not necessarily a bad thing, but it’s a different approach and means that the funding doesn’t necessarily suggest the results one might expect.
> A16Z has certainly had some misses, but one assumes there actually is some product
They don’t need to show a product. It’s been demonstrated that with capital and some skill you can train a foundation model. A16Z has the former. Murati has the latter.
Could somebody explain why there is so much negativity towards Thinking Machines in this thread? I realize that they haven't publicly announced a product yet but presumably the VCs have some idea of what Thinking Machines is building, and they have some pretty significant OpenAI talent on board, including Murati herself. Some amount of skepticism is warranted, of course, but a lot of these comments read as closer to hostility.
> Could somebody explain why there is so much negativity towards Thinking Machines in this thread?
I think it lies at the intersection of a lot of topics where HN comments are hostile: VCs, large fundraising, AI, OpenAI related employees. These are all topics where HN comments are more hostile than pragmatic.
Most of the hostility is just a proxy for “VCs bad” banter.
I have a pretty dim view of VCS, or at least a16z, after seeing some of the awful blockchain tech startups they were throwing money at during the last bubble
Because it looks like VC are spendings ridicoulous amounts of money for hype while many startups with actual products and a product market fit struggle to attract investors because their product is not related to transformers.
I don't think that's fair. The company was founded around 6 months ago and already has a valuation of 12B even though barely anyone outside of AI has heard of them, and I bet almost nobody even here could name their product. It's not that people think she's untalented, but rather, she'd have to be superhuman to justify that by herself.
> but presumably the VCs have some idea of what Thinking Machines is building
you imply that VCs are rational because bet their own money, which in current complicated world probably is not true. VC funds get money from complicated funnel likely including my/your retirement account and country public debt, VC managers likely receive bonuses for closed deals and not long term gains which may materialize in 10 years. So, investing 2B into non-existing product with unclear market fit/team/tech moat smells very strongly.
This is what I expect. Theyre Not hiring even engineers. By the time they have a strategy, plan, hire, and act on that plan they will be behind the curve, and force to use the $ to acquire someone who did.
> "We're excited that in the next couple months we will be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom model," CEO Murati said in a post on the X social media platform.
VCs are betting they'll get out with at least some gains before this bubble bursts. They have literally nothing new, not even in theory. They'll build a smaller-but-not-that-small-still-better-than-you-expect model which will be kind of open for others, which means they'll cut some deal and pretend they did something and that they received something in return.
> "We're excited that in the next couple months we will be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom model," CEO Murati said in a post, opens new tab on the X social media platform.
To be fair the other Thinking Machines has been defunct over 30 years.
That said, what was left of the old one was bought by Sun, which is now owned by Oracle.
I wonder if they still own rights to the name? Not the wisest move to name your new company after something owned by the most famously litigious tech corporation.
Thinking Machines was chosen over the Cray because they had more visual appeal. Sheryl Handler the CEO had (has) a real flair for and it showed; they were neat looking machines
The Cray machines looked more like an airport seating area. Or with later models, obstacles in a laser tag arena. While the Thinking Machines with the moving LEDs looked alive, almost like it was designed to be a character in a movie, which they became.
Fascinating question! I can't find any mention of this seemingly obvious issue.
Here's[1][2] their trademark application from February, which is still "NOT ASSIGNED". Technically it's for their logotype but I imagine it's all the same issue, considering that they include "Computer hardware" in the description of their company (which is exactly what the old one did). This site ominously says that the only action since the filing date was on June 5th, titled "LETTER OF PROTEST EVIDENCE FORWARDED" -- perhaps that's Oracle?
I think this[4] is the trademark for the original's ("Thinking Machines Corporation") trademark logotype, first used in 1987 and defunct ("cancelled"?) by 1999. Another site[5] lists three other "Dead/Cancelled" trademarks owned by the original, and two more recent attempts by randos in 2006 and 2010 that were both shot down.
Technically they're "Thinking Machine Lab Inc."[3], but they're basically always referred to without the "Lab", even to the point of using thinkingmachines.ai as their domain (which, hilariously, doesn't use their trademarked logotype). Another goofy tidbit is that they also filed a trademark for a serif logotype of the words "BEEP BOOP"[6] -- maybe that's their fallback name!
Would be fascinated to hear from anyone familiar with US trademark law on what might be going on, and how we might see what the "LETTER OF PROTEST" is! My layperson understanding would definitely tell me that Oracle would maintain the trademarks, but perhaps they were forced to let them lapse due to lack of use?
I've been slowly building (y'know how it is...) a (one-man...) company filed as "Doering Thinking Machines, LLC" for a few years (named after an old family business, "Doering Machines"), so I'm quite interested to see how this shakes out!
[2] For the love of god, please HN gods, just make these comments markdown. IDK what battle you're fighting but it's a baffling one. The lack of blockquotes is painful, but the lack of inline links is downright diabolical! You have three people now, you can afford the effort ;)
I really hope these seemingly experienced entrepreneurs got the trademark sorted out and locked down before adopting it. Otherwise it may be an expensive lesson.
IANAL but I do know trademarking a logotype is a kind of 'trade dress' that's not the same trademarking the words of the name (even if those words appear\ in the logotype).
At this point, arms-crossed mugshot of "{ex-Open AI} raises $2B" is a meme. Training one epoch of a model the size of the GPT-4 family is easily an 8 figure job. Considering other ops costs, UI and backend dev, securing Everest sized datasets, demos/talks, wages, bonuses etc, this probably gives them around 100 trial and errors, or in other words "Look at me mum, I won 100 chips for the casino!"
And imagine being the ops guy there, about to run a new training batch, but you specified the wrong input path. Puff, the money is gone. Absolutely wild stuff.
Andreessen Horowitz is not gambling with their own money:
https://www.reuters.com/business/finance/andreessen-horowitz...
Moreover, if some banks will fail in the aftermath, they'll be bailed out at the expense of you know whom.
2bn from 16az, post-artificial paradigm auto-correct. this is a market exodus for parting the red sea.
I think it takes a lot of temerity and hubris to look at a 2B raise and assume it's all hype. A16Z has certainly had some misses, but one assumes there actually is some product they're showing behind closed doors that makes this round much more reasonable than it appears from the outside.
If something makes no sense, seems totally crazy, and is being done by a crowd of extremely smart people, you can only assume one of two things: they are actually crazy and frittering away 2B on hype or, just maybe, there's something we're not aware of. If there are only two camps: optimistic and naive or pessimistic and dismissive, I'll choose naive every day of the week.
Anyways, congrats to Thinking Machines and here's hoping they do have something awesome up their sleeve!
> they are actually crazy and frittering away 2B on hype
This assumes that being careless with billions can only ever be crazy.
If you're already set for life, why not gamble (including at completely irrational levels) for even more insane amounts of money when the whole thing is just a crazy house of mirrors. Yes, there's value at the heart, but there's also crazy amounts of money being funneled in, lots of opportunity for chaos, lots of chances for legal rug pulls. All of it inflated even further by a fervor of carelessness for any kind of consequences - things like the stock market are completely removed from any kind of fundamentals.
In a fun house of mirrors, that 2 billion could be 2 cents, or it could be 2 trillion. Buy the ticket and have fun!
> a crowd of extremely smart people
I don't disagree with your broader point, but have spent enough time with enough a16z partners to say they're just people. Not outright stupid, but not extremely smart, either. And their error rate is pretty high.
Which...to some extent is by design. It's part of a VC's job to make bad bets. Sometimes the price of getting into a deal at all is getting in on insane terms, but you still do it because that one investment could return the entire fund. Maybe Thinking Machines is a winner, maybe it's another Clubhouse. We'll see.
I do agree for $2bn, but it’s also well known that US and in particular SV VCs will fund ideas and people, I.e. potential, whereas everywhere else funds results.
It’s that culture that creates some spectacular hits, and a vast number of misses. Not necessarily a bad thing, but it’s a different approach and means that the funding doesn’t necessarily suggest the results one might expect.
> A16Z has certainly had some misses, but one assumes there actually is some product
They don’t need to show a product. It’s been demonstrated that with capital and some skill you can train a foundation model. A16Z has the former. Murati has the latter.
Oh it’s going to get exciting indeed!
Could somebody explain why there is so much negativity towards Thinking Machines in this thread? I realize that they haven't publicly announced a product yet but presumably the VCs have some idea of what Thinking Machines is building, and they have some pretty significant OpenAI talent on board, including Murati herself. Some amount of skepticism is warranted, of course, but a lot of these comments read as closer to hostility.
> Could somebody explain why there is so much negativity towards Thinking Machines in this thread?
I think it lies at the intersection of a lot of topics where HN comments are hostile: VCs, large fundraising, AI, OpenAI related employees. These are all topics where HN comments are more hostile than pragmatic.
Most of the hostility is just a proxy for “VCs bad” banter.
> HN comments are more hostile than pragmatic
In the case of a bubble, hostility is pragmatism.
Of course, nobody knows that it's a bubble, but nobody knows it's not a bubble either.
The math to justify that investment is crazy.
I have a pretty dim view of VCS, or at least a16z, after seeing some of the awful blockchain tech startups they were throwing money at during the last bubble
They were founded 6 months ago, have no product, yet are purported to be worth $12B? I mean ... Come on.
Because it looks like VC are spendings ridicoulous amounts of money for hype while many startups with actual products and a product market fit struggle to attract investors because their product is not related to transformers.
> Could somebody explain why there is so much negativity towards Thinking Machines
> significant OpenAI talent on board, including Murati herself.
because ppl on this website don't consider her a 'talent'.
I don't think that's fair. The company was founded around 6 months ago and already has a valuation of 12B even though barely anyone outside of AI has heard of them, and I bet almost nobody even here could name their product. It's not that people think she's untalented, but rather, she'd have to be superhuman to justify that by herself.
> but presumably the VCs have some idea of what Thinking Machines is building
you imply that VCs are rational because bet their own money, which in current complicated world probably is not true. VC funds get money from complicated funnel likely including my/your retirement account and country public debt, VC managers likely receive bonuses for closed deals and not long term gains which may materialize in 10 years. So, investing 2B into non-existing product with unclear market fit/team/tech moat smells very strongly.
> you imply that VCs are rational because bet their own money, which in current complicated world probably is not true.
Correct. Most VCs are using someone else's money. See Softbank. And making extremely poor judgements on how to use that money.
No one needs another mistral
huh, what happened? I haven't been following Mistral but I recall it was everyone's up and coming darling, no?
It's about to be bought by Apple, so not for long.
I'll take another ten Mistrals thank you.
I guess investors are hoping for another Anthropic, or resurrection of Richard Feynman to help with development.
Does anybody know what they are building yet?
"it wants to build artificial intelligence systems that are safer, more reliable and aimed at a broader number of applications than rivals".
If only I had had that idea, maybe I could have raised $2B.
I am working on a pitch deck for a startup that is going to build language models that aren't just safe and reliable but they are actually CONSCIOUS.
consciousmachines.ai
Such an obvious next step. Conscious, ethical, inclusive machines.
A a supercomputer with 65536 processors and a 16 dimensional hypercube interconnect, in a black monolithic enclosure with red blinkenlights.
Yeah, but no Richard Feynman this time around. And I was just wearing this shirt today! https://www.tamikothiel.com/cm/cm-tshirt.html
Gavin Belson Signature Edition Box 1
why are you giving away the joke
More of the same it seems. How can you deliver a product within 2 months of there is anything novel.
> How can you deliver a product within 2 months of there is anything novel
Uh, run the training script with your thoughtful modifications? They’re not welding together an LLM.
hype
This is what I expect. Theyre Not hiring even engineers. By the time they have a strategy, plan, hire, and act on that plan they will be behind the curve, and force to use the $ to acquire someone who did.
Well this is blatantly false, she linked the career page and I know of people that received offers recently.
They have very strong talent from Meta's FAIR/Pytorch teams as well as a lot of strong people from OAI.
This is wrong. they have strong engineering and a product coming this year
While you are seeing in to the future, can you tell me the Powerball numbers?
It is literally in the article:
> "We're excited that in the next couple months we will be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom model," CEO Murati said in a post on the X social media platform.
Is there a word (like aphorism or metaphor) for someone trying to prove a point and doing the opposite? There just has to be.
"You've just proved my point" is enough.
What are they building?
they have concepts of a product
A big, beautiful... thing.
Wouldn't overlook this company, they manage to bring over a good deal of top talent from OpenAI including John Schulman
VCs are betting they'll get out with at least some gains before this bubble bursts. They have literally nothing new, not even in theory. They'll build a smaller-but-not-that-small-still-better-than-you-expect model which will be kind of open for others, which means they'll cut some deal and pretend they did something and that they received something in return.
> They have literally nothing new, not even in theory
They haven’t built anything yet. It’s a bit premature to call out the non-existing product as insufficiently novel.
Well, if you have to pay $200 million to a top AI engineer, $2B is a very short runway for a startup.
I need to get around to betting against this on prediction markets soon.
> "We're excited that in the next couple months we will be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom model," CEO Murati said in a post, opens new tab on the X social media platform.
We need to jack up interest rates again.
hahaha
So novel, they couldn't even come up with an original name.
To be fair the other Thinking Machines has been defunct over 30 years.
That said, what was left of the old one was bought by Sun, which is now owned by Oracle.
I wonder if they still own rights to the name? Not the wisest move to name your new company after something owned by the most famously litigious tech corporation.
Yet the Jurassic Park franchise is just as strong as ever.
Thinking Machines CM-5 in Jurassic Park (1993):
https://www.starringthecomputer.com/appearance.html?f=11&c=1...
Jurassic World Rebirth passes $500 M in revenue:
https://www.koimoi.com/box-office/jurassic-world-rebirth-wor...
Fun fact, in the novel the computers were Cray X-MPs (sadly Cray is now semi-defunct, since they were bought and merged into HPE) [1].
[1] https://cray-history.net/2023/08/20/cray-systems-in-popular-...
Thinking Machines was chosen over the Cray because they had more visual appeal. Sheryl Handler the CEO had (has) a real flair for and it showed; they were neat looking machines
The Cray machines looked more like an airport seating area. Or with later models, obstacles in a laser tag arena. While the Thinking Machines with the moving LEDs looked alive, almost like it was designed to be a character in a movie, which they became.
Maybe the new company should consider something like this for the next investor show and tell
https://rodyne.com/?p=1674
what? Cray machines were pretty rad looking in their own way.
They were, but they didn't blink.
One would imagine they already thought of that. There are at least two people at a16z old enough to know about the original TM.
Fascinating question! I can't find any mention of this seemingly obvious issue.
Here's[1][2] their trademark application from February, which is still "NOT ASSIGNED". Technically it's for their logotype but I imagine it's all the same issue, considering that they include "Computer hardware" in the description of their company (which is exactly what the old one did). This site ominously says that the only action since the filing date was on June 5th, titled "LETTER OF PROTEST EVIDENCE FORWARDED" -- perhaps that's Oracle?
I think this[4] is the trademark for the original's ("Thinking Machines Corporation") trademark logotype, first used in 1987 and defunct ("cancelled"?) by 1999. Another site[5] lists three other "Dead/Cancelled" trademarks owned by the original, and two more recent attempts by randos in 2006 and 2010 that were both shot down.
Technically they're "Thinking Machine Lab Inc."[3], but they're basically always referred to without the "Lab", even to the point of using thinkingmachines.ai as their domain (which, hilariously, doesn't use their trademarked logotype). Another goofy tidbit is that they also filed a trademark for a serif logotype of the words "BEEP BOOP"[6] -- maybe that's their fallback name!
Would be fascinated to hear from anyone familiar with US trademark law on what might be going on, and how we might see what the "LETTER OF PROTEST" is! My layperson understanding would definitely tell me that Oracle would maintain the trademarks, but perhaps they were forced to let them lapse due to lack of use?
I've been slowly building (y'know how it is...) a (one-man...) company filed as "Doering Thinking Machines, LLC" for a few years (named after an old family business, "Doering Machines"), so I'm quite interested to see how this shakes out!
[1] https://furm.com/trademarks/thinking-machines-99054776
[2] For the love of god, please HN gods, just make these comments markdown. IDK what battle you're fighting but it's a baffling one. The lack of blockquotes is painful, but the lack of inline links is downright diabolical! You have three people now, you can afford the effort ;)
[3] https://trademarks.justia.com/741/37/thinking-machines-74137...
[4] https://en.wikipedia.org/wiki/Thinking_Machines_Lab
[5] https://uspto.report/TM/99051772
[6] https://trademarks.justia.com/990/71/beep-99071391.html
I really hope these seemingly experienced entrepreneurs got the trademark sorted out and locked down before adopting it. Otherwise it may be an expensive lesson.
IANAL but I do know trademarking a logotype is a kind of 'trade dress' that's not the same trademarking the words of the name (even if those words appear\ in the logotype).
I guess "random and pleasing" is appropriate for both CM-5s and LLMs.
Maybe it's a throwback fashion statement.
Bet they've never heard of it