Interested in anyone who knows about this company (they have a lot of ML hiring listings right now). Basically wondering if it’s worth exploring more. They have apparently raised $240mln and are a unicorn so this is an important topic.
Here is the summary of some red flags I have found:
-
Founders have no ML background
-
Zero released product after several years despite huge funding.
-
TC article says founded 2021, but every listing claims it is YC2017. One of the founders did a recruiting service from YC 2017, but Imbue is a totally unrelated company with a different founding group so claiming YC affiliation seems dubious/unethical. YC is also not named as an investor in the company anywhere on their website.
-
No-one I have spoken to has ever worked with them/ heard of them outside press
-
Claim to have raised $240mln, but not clear who from. Named investors on recent $200mln round are “Astera Institute, NVIDIA, Cruise CEO Kyle Vogt, Notion co-founder Simon Last and others” which seems like a random group and there is no named lead. As a matter of fact, there is not a single VC named anywhere in their marketing as an investor which would be very unusual for a SF AI startup that has raised >$200mln. Crunchbase also has no VCs/institutional investors recorded. I couldn’t find any investor that I would expect to have done significant due diligence. A month after they raised $200mln, they prominently announced that they raised another $12mln from Amazon Alexa and Eric Schmidt - this seems like a pure publicity stunt to me. This announcement of an extra 5% in round funding got a bunch of press articles about it also. They push the press angle incredibly hard.
-
Their CEO has been in the press claiming that they buy their own chips, and they also claim to have a 10k H100 stack, but they literally haven’t raised enough funding to buy that many H100s. Maybe they are renting and all the press about buying is just press?
-
The press/publicity to product ratio is extremely high. Dozens of articles in major publications with nothing released and no claimed customers.
-
Their site is full of gushy statements with no detail like “We pretrain very large (>100B parameter) models, optimized for internal reasoning benchmarks.” (internal reasoning benchmarks is a great way to pretend you have competitive models!) and “Our north star: truly personal computers that give us freedom, dignity, and agency to do what we love”. As far as I can tell, Imbue has never released even a 3B or 7B open source model like a bunch of similar startups have.
-
Their “Our Work” blog that should showcase their research and products contains 4 articles from 2023. 2 of them are opinion pieces on LLM ethics that contain no technical details, 1 is a niche piece on RL in hyperparameter search, and 1 discusses SSL on ResNet50 - a 23mln param CNN that is 5,000 times smaller than the LLMs they are apparently training.
Of course red flags are just that - suggestions of a problem not evidence. Inflection is in a similar boat on many of the above points - their only public product is the inferior Pi model - but at least they have serious industry heavyweights in their management and research so I give them the benefit of the doubt.
So, to those in the know - is this a real company? Has it duped a bunch of rich individuals? Is it worth spending more time on?
It’s a real firm. They’re associated with Ycombinator.
The direction of achieving robust reasoning in LLMs is both reasonable and likely to be achievable. It probably isn’t super easy, but lots of people are thinking about ways of achieving it and it’s an active area of research.
I’m not sure whether I’d invest in it, but it seems fine. It’s a reasonable direction and I’m sure there’s a way for people to be productive while working in that direction.