wsgeorge

19 ms·

181.
by wsgeorge 4y ago
This is such an interesting use case. Rather can compressing existing information, LLMs "model" it, so they can recreate what they've learned. For all Elon's concerns, these might be a great way of seeding the universe w
182.
by wsgeorge 4y ago
That's a remarkable list of signatories.
183.
by wsgeorge 4y ago
Looks like really positive progress! Aiming to match (and surpass) GPT-4 is a great goal, iirc. I'm not sure there's any open source model that matches even 3.5 Turbo on benchmarks. That said, I would've wished the demo allow
184.
by wsgeorge 4y ago
Fascinating read. Historical documents have a way of throwing you back into the times to see the world as the authors saw it. There's so much to wonder about this text, dated 25th Feb 1945, in light of the hopelessness of the German si
185.
by wsgeorge 4y ago
> In many ways software is the most egalitarian product in history, Indeed, and that's kind of my point. Even in the most egalitarian product in history, a bit of a wealth gap exists. It's inescapable. That $60 game may not cos
186.
by wsgeorge 4y ago
I see. Hadn't considered this. Yes, I see how that might be a concern. What I think is important in this AI Spring is that we make it possible for people to run their own models on their own hardware too, without having to submit anyth
187.
by wsgeorge 4y ago
Bingo! I'm lifting this idea out of Julian Jayne's theory of consciousness :D https://www.julianjaynes.org/about/about-jaynes-theory/overv...
188.
by wsgeorge 4y ago
Ask yourself the same question but for web services, games, etc. A wealth gap will exist wrt access to better AI.
189.
by wsgeorge 4y ago
An even more fun experiment will be to have two models running in perpetuity, constantly talking to each other, but constructed to act as though they were two sides of the same model.
190.
by wsgeorge 4y ago
> You can't know if the answer was real or stochastic parroting. I feel like at some point we will have to come to terms with the fact that we could say the same for humans, and we will have to either accept or reject by fiat that a
191.
by wsgeorge 4y ago
> Do you know if that hardware is capable of running LLM or is that optimized for something else? I do! Checkout llama.cpp [0] and related projects. The game changer with this was Gerganov's ggml format that makes inference on CPUs
192.
by wsgeorge 4y ago
I think GPT-3 is used as a benchmark for performance, so saying a model is on par with GPT-3 should give you an idea of what you can get out of it. IIRC most open source models to date - including the semi-open LLaMAs - have GPT-3-like perf
193.
by wsgeorge 4y ago
I guess it's something. It still goes to show how far open models are behind the proprietary SOTA.
194.
by wsgeorge 4y ago
I believe this has been extended to mean "on device", which is interesting. See Gerganov's article on Github [0]. I wrote about this here [1] where I made a contrast between the core and the edge. I think the term maps well t
195.
by wsgeorge 4y ago
If Gary is patient we could get there soon enough. The exciting thing about these debates (insufferable as I think they are because they're kind of a distraction from the real potential and threats of LLM-based systems) is seeing how m
196.
by wsgeorge 4y ago
Even better, run locally. That should be the goal.
197.
by wsgeorge 4y ago
By "grounded", I mean the promises of crypto vs the promises of AI. IIRC, crypto was a little too idealistic in the promise of providing a digital currency without needing a central, trusted authority to back it. From my perspecti
198.
by wsgeorge 4y ago
Crypto wasn't good enough as a currency, so what was coming in terms of that use case didn't pan out en masse. But it did work to an extent for some people. Crypto's value fluctuation made more sense as a speculative asset cl
199.
by wsgeorge 4y ago
AFAIK it's still an active area of research, and evidence from Meta AI [0] suggests that size and quality of data can let smaller (not necessarily less sophisticated) models do amazing things. But a lot of the advancements we're s
200.
by wsgeorge 4y ago
> I quit posting articles about that to HN because they never get a second vote. Just because your links don't make the frontpage doesn't mean you should stop sharing. HN is a hard nut to crack to get something to get the atten
201.
by wsgeorge 4y ago
Depending on how mature AI will be, people's work will involve telling AIs what they want it to do, rather than doing the work themselves. I wrote about this here [0] because it seems like a logical step after technology improves any l
202.
Ask HN: Is it too early to start comparing LLM prompting to mental priming?
1 points by wsgeorge 4y ago|1 comments
203.
by wsgeorge 4y ago
> Nothing to maintain. Never say never...
204.
When production is commoditized and scaled by AI, only ideas will be king(medium.com)
2 points by wsgeorge 4y ago|0 comments
205.
by wsgeorge 4y ago
The most interesting thing about this article is the claim that GPT-4 has 1 trillion parameters. Microsoft's recent GPT-4 paper [0] hints at the "unprecedented scale of compute and data" used to train the model. What else do
206.
by wsgeorge 4y ago
2012 feels like year zero for a lot of the change we're experiencing now.
207.
by wsgeorge 4y ago
It's a bit of a shame that open source LLMs lag so far behind OpenAI. But it's more to their credit: GPT-4 outclasses the competition from the published benchmarks. Unless a team has been working on a competitor for a while, we&#x
208.
Ask HN: People here seem to really dislike Lex Fridman. What's the backstory?
4 points by wsgeorge 4y ago|9 comments
209.
by wsgeorge 4y ago
I've learned to not trust external hard drives, so my most important work is backed up locally and on redundant cloud accounts. However, I've heard too many horror stories about WD. Is this a thing? Also, a hard drive ratings site
210.
by wsgeorge 4y ago
> Is it even possible to coordinate something like that. Hugging Face and other collaborators led the creation of BLOOM. Given how hot the competition is, I imagine some team will take advantage of this soon...

More ›