golol

23 ms·

331.
by golol 3y ago
At the end the article presents a clear example where a model is trained on A is B and infers B is A. That is the contribution of this article and it's very valuable.
332.
by golol 3y ago
Excellent blog post. The only thing which I wish was there was an explanation of how promot + completion training would be any different to pure completion training. From my understanding oftransformers, it ahould be the same.
333.
by golol 3y ago
Just to be clear: You are testing with GPT-4 right?
334.
by golol 3y ago
Nice interview. I think in the near future human art will be seen like "handmade" or "bio" products, used when money is available and quality is desired. There is currently a large market for art that is purely functiona
335.
by golol 3y ago
I met my SO on Omegle...
336.
Ask HN: Where Are the Text+video Models
1 points by golol 3y ago|1 comments
337.
by golol 3y ago
I think "GPTs" is peak bitter lesson
338.
by golol 3y ago
Yes, agents don't really work yet, I agree.
339.
by golol 3y ago
Well, I would say that when people at the moment say agent they mean what you call autonomous agent. Browsing or code interpreter is just LLM + tool use. I think there is a really large difference in quality between a system that just inter
340.
by golol 3y ago
Not even. An agent should primarily operate in some kind of self-prompting loop. Afaik xou can not specify complex and branching looping behaviors for GPTs.
341.
by golol 3y ago
Any entity should be allowed to read, process and update its internal state using published media. Only your productions should be judged based on copyright law.
342.
by golol 3y ago
SGD in high dimensions is good at finding efficient, generalizing models. Naturally, it is more efficient to learn a representation of the abstract concept of a table rather than learn representations for the hundreds of different words for
343.
by golol 3y ago
In-comtext learning exists though.
344.
by golol 3y ago
Time series prediction is always about using the particular features of your distribution of time series. In standard time series prediction the features of the distribution are mostly things like "periodic patterns are continued"
345.
by golol 3y ago
> This doesn’t improve our national security at all. You said yourself that other nations can and will make this too. Avoiding a decrease of national security is the same as improving national security.
346.
by golol 3y ago
> Drones and robots made attacks easier, did that help national security? Yes because the enemy will have them eventually regardless of whether you develop them. Call it a Prisoner's Dilemma if you will, but it's the reality. B
347.
by golol 3y ago
European security is national security via geopolitics. Being a global superpower has many benefits for US citizens.
348.
by golol 3y ago
There is little difference between offensive and defensive capabilities. It's all about destroying your opponent. This helps national security because it is an additional and advanced means of destruction that the enemy might not have.
349.
by golol 3y ago
Classic off by 1 :)
350.
by golol 3y ago
The author says: "Reasoning, the type you get from scaling these models to get larger, doesn’t matter for 85% of use cases. Researchers love sharing that their 200B param model can solve challenging math problems or build a website fro
351.
by golol 3y ago
The simple trick has already been discovered See this tweet. https://twitter.com/GrantSlatton/status/1703913578036904431 I have tested it myself.
352.
by golol 3y ago
Just to make it clear: the database doesn't have to contain exclusively top rated games. It can be 99% low ELO games, as long as there are suffiiciently many high ELO games (in absolute number), the LLM is large enough, and there is su
353.
by golol 3y ago
That is extremely impressive for a model that can do literally millions of tasks. It is not trained to play chess. It can write stories, code, recipes, franslate text, create JSON AND play chess. Don't you see how amazing this is compa
354.
by golol 3y ago
This is not true anymore, see this tweet. https://twitter.com/GrantSlatton/status/1703913578036904431 I have tested it myself.
355.
by golol 3y ago
This is not true anymore. It has been discovered that gpt-3.5-instruct when prompeted with purely PGN notation.playes at >1700 Elo and makes no/little illegal moves.
356.
by golol 3y ago
It has learned an algorithm that plays chess purely by reading chess logs. It can play at different skill levels depending on the surrounding natural language How is that not understanding? By the way, would you say that Stockfish 15 unders
357.
by golol 3y ago
Except they can play chess better than 90% of humans... Just by luck of course..No understanding.
358.
by golol 3y ago
My point is that we have to try, because there are still big issues in this world we should try to fix. We're just gonna have to try and make it work.
359.
by golol 3y ago
> Most researchers don't take the LessWrong in-culture talk seriously Yes but politicians do, for some reason. AI Safety has become a meaningless term, because it is so broad it ranges from "no slurs please" over "div
360.
by golol 3y ago
Look, there is no AI safety advancement without AI capability advancement. I think we van learn fuck all about AI safety if we don't try to actually build those AIs, carefully, and play around with them. AI safety is not an actual fiel

More ›