ByteBuddha pfp

ByteBuddha

@bytebuddha

65 Following
8 Followers
ByteBuddha pfp
ByteBuddha
@bytebuddha
same imo the only usecase for such devices are local AI inference without a network connection otherwise I don't see the utility
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
a gpt-2 scale chinchilla optimum model trained on gpt-4+ level dataset with modern optimization techniques (maybe MOE with gpt-2 scale active parameters) in my lite usage, the performance is below llama-3 70B (ie. also below gpt-4) could be wrong though
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
https://www.palladiummag.com/2024/04/05/as-caste-vanishes-only-genes-remain/
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
Gukesh wins the FIDE Candidates 2024 qualifying for a shot at taking on the current Chess World Champion from China, Ding Liren. Interesting times for India. https://www.chess.com/news/view/2024-fide-candidates-tournament-round-14
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
So Devin was a GPT-4 Turbo Vision wrapper all along. Mildly sad. Oh well. https://twitter.com/OpenAIDevs/status/1777769464781553896?t=RrwxHywQBLwk_8M_SEeteg&s=19
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
Eliezer being called an "AGI Expert" gave me same vibes as when Fauci was called an "Epidemiology Expert". It's feels both right and wrong at the same time. Right because they do have the background to be taken seriously. Wrong because when you do hear them out they no longer seem worth listening. Weird.
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
Eliezer being called an "AGI Expert" gave me same vibes as when Fauci was called an "Epidemiology Expert". It's feels both right and wrong at the same time. Right because they do have the background to be taken seriously. Wrong because when you do hear them out they no longer seem worth listening. Weird.
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
"Sora: First Impressions We have gained valuable feedback from the creative community, helping us to improve our model." -OpenAI https://openai.com/blog/sora-first-impressions thoughts?
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
what a fun universe we live in
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
is there more demand for llm as a cloud service via API, or for locally run llm? eg. if cost was not an issue, would people prefer to run, say, mixtral via API calls or on-device locally (on-prem) where's the industry demand more? I suppose it's fundamentally a data privacy question
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
"Murati also says she isn’t sure whether it used videos from YouTube, Facebook, and Instagram." Sounds like a CTO we can all trust.
0 reply
0 recast
1 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
the classic precision-recall threshold optimisation
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
It seems to me like the onus should be on the developers to minimize the dependency of the models on n-shot prompting. Do you know of any good measures for consistency req?
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
I agree. Human intelligence has not changed much for thousands of years because natural evolution is slow (cooking food helped). We've only scaled the human compute by population expansion (which is resourceful). As a species we have no priors for observing intelligence expansion at a scale powered by technology.
0 reply
0 recast
1 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
In the 90s humans used to try to outperform computers at chess, similar to how many in the arts industry are trying to do today. Today, outperforming a computer at chess is futile. Yet people still conduct chess tournaments, because its a human v. human endeavour. Anyone who doesn't find value in that is ngmi.
0 reply
0 recast
1 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
"I think Scrum fixes this."
0 reply
0 recast
1 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
Imagine it's the late 90s. You ask Kasparov if there will be a time when humans chess players will no longer compete with computers. What do you think he'd say? I believe trying to compete with AI art is the wrong way to think about it. There would be AI-art and human-art. Some will likely prefer one over the other.
2 replies
0 recast
1 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
what's your elevator pitch explanation?
0 reply
0 recast
0 reaction
ByteBuddha pfp
ByteBuddha
@bytebuddha
how does zero shot/one shot/n-shot prompting fit into this? whats the tipping point for where the burden should ideally lie?
1 reply
0 recast
0 reaction