emadm

9 ms·

91.
by emadm 3y ago
Its the standard nomenclature to use when you have graduated from Oxford with 7 years post matriculation https://en.wikipedia.org/wiki/Master_of_Arts_(Oxford,_Cambri... I don't get the issue when this is a standar
92.
by emadm 3y ago
We make private versions of the models and serve them at scale cheap as we have loads of fast chip access when the market has run out. Had a record month It’s known as an open core business model https://twitter.com/emostaqu
93.
by emadm 3y ago
The spy thing was puzzling then realised it was likely because it my previous work on counter extremism some elements of which were public and they didn’t bother checking https://twitter.com/emostaque/status/168937
94.
by emadm 3y ago
He wanted out completely and no association after first project failed https://twitter.com/emostaque/status/1680774535342358528?s=4...
95.
by emadm 3y ago
It was a weird allegation when I paid the postage and got the MA from Oxford a couple of weeks later https://twitter.com/emostaque/status/1682091613278072832?s=4... Told them that and more https://emad.
96.
by emadm 3y ago
Intel Gaudi2 is actually really good, better than A100s and cheaper (including interconnect, 2.15 Tb/s) CUDA isn't that important any more. The weird one was that Aurora at 2 exaflops using.. Intel Max?
97.
by emadm 3y ago
This is true, Tesla took most of the remainder. We took 30 MW outside the US but also some inside the US
98.
by emadm 3y ago
Also missed in the post is fp8 is really much more efficient The H100s are actually very good for inference..
99.
by emadm 3y ago
Gaudi2 and Inferentia2 are both good. We train on A100s, TPUs and... other things now.
100.
by emadm 3y ago
Don’t need lock in just support. Business model is called open core hundred of billions in market cap here as most folk don’t want to build/train their own models and will pay for support and help https://en.m.wikipedia.org&
101.
by emadm 3y ago
Make models usable is really valuable, for Stability AI I discussed business models with Sam Lessin here: https://www.youtube.com/watch?v=mOOYJONenWU but basically the edge is data and distribution given how widely used thi
102.
by emadm 3y ago
Yes that would be the case
103.
by emadm 3y ago
It outperforms chat tuned LLaMA handily https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderb...
104.
by emadm 3y ago
It beats GPT 3.5 in some benchmarks, the first open model to do so I believe. Versions being worked on now will do much better. GPT 4 is far better and will likely not be beaten by any current open models and approaches but maybe an ensembl
105.
by emadm 3y ago
Not yet, update soon on that. Just had to change tack with LLaMA 2, we trained and released OpenLLaMA 13b in the meantime.
106.
by emadm 3y ago
https://www.cultofmac.com/475927/ron-wayne-quits-apple/
107.
by emadm 3y ago
I did apologise and apologise again recompense when payment was late during a difficult time in the scale up and stepped in with my own money. I agree Eric is bright and so wanted to support him through an internship where he had lots of fr
108.
by emadm 3y ago
Yes if there aren’t good open models available then it’s massively centralising and will lead to a massive panopticon
109.
by emadm 3y ago
Eric didn’t manage to do his internship but we still offered to pay for the whole thing. We missed one invoice and accounts asked him to send details so he could settle but instead he decided to a month later say that stuff to Forbes then s
110.
by emadm 3y ago
…but we release our stuff open source and are one of the largest supporters of open source AI out there from funding https://github.com/lucidrains to compute grants for awesome stuff like OpenFlamingo 2 https://t
111.
by emadm 3y ago
We actually have a blog post coming on specifically this from AWS, at the time it was the 8th largest public A100 cluster and we covered it live again while we were preseed well before the funding round https://twitter.com/e
112.
by emadm 3y ago
Yes we told them this and other things multiple times You don’t actually get your bachelors or masters without either going to ceremony or asking for it in absentia so like I’m going to get them in a few weeks, do a photoshoot
113.
by emadm 3y ago
Yep
114.
by emadm 3y ago
Thus there is a market on creating open, auditable models of every media type, with sectoral and national variants. Models are like talented grads that occasionally go off their meds, proprietary black box models are like hiring them from M
115.
by emadm 3y ago
Cheers. I have written about our business model many times here is a thread https://twitter.com/emostaque/status/1649152422634221593?s=4... Also did a recent podcast on various AI business models with Sam Lessin t
116.
by emadm 3y ago
I have seen the orders and there simply isn’t enough looking at inference demand
117.
by emadm 3y ago
That was an awful piece of journalism where they ignored multiple factual corrections after spending days with me. Even met the family and made fun of my work on my sons autism which shows what they are like. How many companies preseed get
118.
by emadm 3y ago
Opposite actually.
119.
by emadm 3y ago
It means I have exceptionally high confidence that this will be the biggest thing to hit the economy, society and markets in the last hundred years for good and ill. Against this there is minimal risk in obtaining large scale supercompute w
120.
by emadm 3y ago
20b done

More ›