cardine

6 ms·

1.
by cardine 7mo ago
I think this risk is much lower in a world where there are lots of different model owners competing with each other, which is how it appears to be playing out.
2.
by cardine 8mo ago
Nomi.ai | https://nomi.ai | Senior Machine Learning Engineer | Remote (Global) | Full-time | $150k–$250k + equity At Nomi, we're building AI companions that form deeply meaningful, humanlike relationships and immersive role
3.
by cardine 8mo ago
> make sense and are reliable If you can figure out how to create benchmarks that make sense, are reliable, correlate strongly to business goals, and don't get immediately saturated or contorted once known, you are well on your way
4.
by cardine 1y ago
If they had stayed silent since GPT-4, nobody would care what OpenAI was releasing as they would have become completely irrelevant compared to Gemini/Claude.
5.
by cardine 1y ago
Nomi.ai | Senior Machine Learning Engineer | Remote (Global) | Full-time | $150k–$250k + equity At Nomi, we're building AI companions that form deeply meaningful, humanlike relationships and immersive roleplaying experiences. With over
6.
by cardine 1y ago
Nomi.ai | Senior Machine Learning Engineer | Remote (Global) | Full-time | $150k–$250k + equity At Nomi, we're building AI companions that form deeply meaningful, humanlike relationships and immersive roleplaying experiences. With over
7.
by cardine 1y ago
Founder/CEO of Nomi here. The story in question was someone who intentionally jailbroke our LLM for a misleading news story. The same things done in that article can be done for ChatGPT, Gemini, etc. and since the article was published
8.
by cardine 3y ago
I wonder if Sam knew he was going to lose this power struggle and then started working on an exit plan with people loyal to him behind the boards back. The board then finds out and rushes to kick him out ASAP to stop him from using company
9.
by cardine 4y ago
You might not care but that doesn't make calling them out for reneging on their original mission a trivial and unsubstantial critique.
10.
by cardine 4y ago
In addition to very open publishing, Google recently released Flan-UL2 open source which is an order of magnitude more impressive than anything OpenAI has ever open sourced. I agree, it is a bizarre world where the "organization that l
11.
by cardine 4y ago
OpenAI didn't pick that name arbitrarily. Here was their manifesto when they first started: https://openai.com/blog/introducing-openai > OpenAI is a non-profit artificial intelligence research company. Our goal
12.
by cardine 4y ago
> Given both the competitive landscape and the safety implications of large-scale models like GPT-4, this report contains no further details about the architecture (including model size), hardware, training compute, dataset construction,
13.
by cardine 4y ago
I don't think the person you were responding to was claiming that. The brain plausibly having something akin to a language model doesn't imply that building or studying language models will unlock a better understanding of the bra
14.
by cardine 4y ago
And yet there are still no publicly available models that could actually compete with ChatGPT. I'm not even talking about RLHF (although data like that is also a huge moat) - just simple things like larger context sizes. There are stil
15.
by cardine 4y ago
This is a very cool idea. We are doing something similar except we are also predicting the nodes. In the end, the winning combination will likely be doing both. There will be a predicted graph structure which serves as a high level guide to
16.
by cardine 4y ago
As mentioned in another comment, the contract has very clear language not to share it - likely because they are offering different prices to different companies. So I don't feel comfortable sharing any specifics, especially since this
17.
by cardine 4y ago
The contract has very clear language not to share it - likely because they are offering different prices to different companies. (And as p1esk mentioned, there is no way you are getting H100s for under $100k).
18.
by cardine 4y ago
> I suppose if I had a 7 digit budget I could get a better deal. We got our "deal" when buying just a single server and have since bought many more with the same provider. We didn't spend 7 figures all at once, we did it p
19.
by cardine 4y ago
I know how much we paid and it is substantially less than what you were quoted - very likely from one of the 12 providers you contacted. It is likely you just didn't realize how much margin these providers have and did not negotiate en
20.
by cardine 4y ago
I'd suggest finding a cheaper vendor if that is the lowest price you can get for an 8xA100 server. We spend a lot on both and colo our servers so I've definitely done the math!
21.
by cardine 4y ago
This is not true - the break even period is closer to 6-7 months.
22.
by cardine 4y ago
Don't mind at all :)
23.
by cardine 4y ago
Glimpse.ai | Senior/Lead Machine Learning Engineer | Remote | Full-time | VISA Sponsorship Available | $200k - $350k + Benefits + Equity | https://www.glimpse.ai Glimpse.ai is a profitable, stable, and growing artificial in
24.
by cardine 4y ago
They've largely stopped publishing.
25.
by cardine 4y ago
A lot of computation is offloaded to the CPU, such as gradients and optimizer states. You are right though that quite a bit of computation is still done on the GPU.
26.
by cardine 4y ago
Offloading is when the computation is done on the CPU instead of the GPU. DeepSpeed is an example of this.
27.
by cardine 4y ago
Someone has already created a proof of concept for this: https://incoherency.co.uk/blog/stories/sockfish.html
28.
by cardine 4y ago
Glimpse.ai | Senior Machine Learning Engineer | Maryland or Remote | Full-time | VISA Sponsorship Available | $150k - $300k + Benefits + Equity | https://www.glimpse.ai Glimpse.ai is a profitable, stable, and growing artificial
29.
by cardine 4y ago
> I can only guess either safety from misuse or leveraging it for money. The former is being used as justification for the latter.
30.
by cardine 4y ago
We have different servers for each. But the split is usually 80%/20% for inference/training. As our product grows in usage the 80% number is steadily increasing. That isn't because we aren't training that often - we are

More ›