Palmik

21 ms·

331.
Elon Musk's X AI startup seeks to raise $1B in equity(theguardian.com)
2 points by Palmik 3y ago|0 comments
332.
by Palmik 3y ago
I see, my bad!
333.
by Palmik 3y ago
The parent said "my company that spends many millions per year".
334.
by Palmik 3y ago
The table is *highly* misleading. It uses different methodologies all over the place. For MMLU, it highlights the CoT @ 32 result, where Ultra beats GPT4, but it loses to GPT4 with 5-shot, for example. For GSM8K it uses Maj1@32 for Ultra an
335.
by Palmik 3y ago
Curious that the metrics [1] of Gemini Ultra (not released yet?) vs GPT4 are for some tasks computed based on "CoT @ 32", for some "5-shot", for some "10-shot", for some "4-shot", for some "0-sho
336.
GPT-Fast: Simple and efficient GPT inference in <1000 LOC of Python(github.com)
1 points by Palmik 3y ago|1 comments
337.
by Palmik 3y ago
In ELO like match-making, you typically pair together people such that they are likely to have 50% chance to win. Therefore as the OP says, filtering down to people with high (60+%) life-time win-rate creates some sort of (interesting) bias
338.
10 years of FAIR: Decade of advancing the through open research(ai.meta.com)
2 points by Palmik 3y ago|0 comments
339.
by Palmik 3y ago
This is a really impressive result -- a model that rivals (and I would say is better than) SDXL that cost only low five figures to train. It is really good at following complex prompts, you can find some examples here https://www
340.
PixArt-α: Fast Training of Diffusion Transformer for Text-to-Image Synthetis(huggingface.co)
3 points by Palmik 3y ago|1 comments
341.
by Palmik 3y ago
For many, it's much easier to understand than it is to memorize.
342.
Elon Musk on Advertisers, Trust and the "Wild Storm" in His Mind [video](youtube.com)
1 points by Palmik 3y ago|0 comments
343.
DocDecoder: A Chrome extension that summarises web policies with GPT-4(chrome.google.com)
3 points by Palmik 3y ago|0 comments
344.
by Palmik 3y ago
Posting this because people might find it interesting, given the Q* project at OpenAI :) https://news.ycombinator.com/item?id=38386487 This is a relatively old paper, but I always found it fascinating. In summary, they trai
345.
Deal or No Deal? End-to-End Learning for Negotiation Dialogues(research.facebook.com)
2 points by Palmik 3y ago|1 comments
346.
We have reached an agreement for Sam to return to OpenAI as CEO(twitter.com)
4 points by Palmik 3y ago|0 comments
347.
by Palmik 3y ago
Recently he also said “There’s no outcome where this organization [OpenAI] is one of the big five technology companies”.
348.
by Palmik 3y ago
Duplicate: https://news.ycombinator.com/item?id=38347501
349.
I deeply regret my participation in the board's actions(twitter.com)
619 points by Palmik 3y ago|412 comments
350.
by Palmik 3y ago
I guess he would prefer is the existing incumbents got even larger, or if his competitor to ChatGPT (Poe) could capture significant fraction of the market.
351.
by Palmik 3y ago
Accelerating research on increasing reasoning / intelligence capabilities of AI? That is opposite of what Emmett seemed to advocate recently, so it would be strange to pick him as the CEO if they really wanted to accelerate that.
352.
by Palmik 3y ago
The default winners in this race, in the medium-term horizon (~18 months), are the existing incumbents: Alphabet, Microsoft, Meta. OpenAI was positioned to be able to compete. I am not sure a diluted OpenAI will be.
353.
by Palmik 3y ago
Emmett is a self-proclaimed AI doomer. Given some of his Tweets from the past, it's likely the board indeed wants someone who will takes things slower: https://archive.is/tuf3s https://archive.is/sNvvs
354.
by Palmik 3y ago
Dupe: https://news.ycombinator.com/item?id=38287043
355.
by Palmik 3y ago
The gap between OpenAI and Google is huge, especially taking into account the low answer rate for PaLM 2. This is worrying.
356.
Notes on VLLM v.s. DeepSpeed-FastGen(blog.vllm.ai)
3 points by Palmik 3y ago|0 comments
357.
by Palmik 3y ago
Companies try to optimize profit for their shareholders, which includes paying as little as the workers will bear (or less, as in some cases people need to work to survive). Similarly, I think people have the right to optimize their time an
358.
Journal of Controversial Ideas(journalofcontroversialideas.org)
1 points by Palmik 3y ago|0 comments
359.
by Palmik 3y ago
Just a guess: (1) A lot of "me too" pre-training that should really be fine tuning; (2) lot of proprietary models whose weights will never be released (or even used :)).
360.
Google's AI-powered search experience expands globally(techcrunch.com)
2 points by Palmik 3y ago|0 comments

More ›