hallqv

Why would we be approaching a plateau?

RLHF is not part of LLaMa pretraining, or pretraning of any other models for that matter. RLHF comes after pretraining. https://twitter.com/Jeande_d/status/1661833563069620247/phot...

Great news for most people at HN and society as a whole. Mating and sex is a highly motivational force for the individual and earning a high income is beneficial for society, so it's great that individual incetives are aligned with societal ones.

Multiple court reaffirms in favor of scraping, including by the SC, seems more like verdant to me..

What rulings are you referring to? Been reaffirmed that scraping LinkedIn is legal multiple times, even by Supreme Court. https://techcrunch.com/2022/04/18/web-scraping-legal-court/

Why not use Elasticsearch/Opensearch? Way more battle-tested and has many more features for text data. Also uses the same vector indexing algos.

Congrats spacex team!!! Keep inspiring the rest of us!!!

Nice distinction, I'm definitely a street fighter engineer myself, and I often struggle to understand martial arts engineers. This article helps with that in a weird kind of visceral way.

I would cut costs aggressively to remove the need for additional financing. That is your ticket to freedom and what is best long-term for the company. As CEO you have full rights & responsibility to do this, even if it will wreak havoc in the short term.

I think him buying twitter was great. He is lightyears better than previous owners & management. He is also running two other companies doing extraordinary things in the world.

Of all the people you can hate on in the world, hating on Elon is to me very odd. Tall poppy syndrome I guess...

Disagree, Elon has high credibility IMO, certainly higher than an anonymous source with incentives to slander. Not saying he is perfect but given the scrutiny he is under and how much he is “in the arena” pushing the envelope.

Elon is furthermore a software engineer (he wrote ALOT of code at zip2 & paypal) so your argument falls flat in that regard.

Elon spoke about this in the Twitter earnings call https://www.youtube.com/live/6bgCqskBmtk?feature=share

Anonymous Twitter sources have zero credibility since a good portion of the employees still want to oust Elon.

Wrong. Elon noticed his tweets had low engagement relative to his follower count compared to other accounts.

After some digging in the code base it turned out that there was a de-boosting factor based on the absolute numbers of blocks your account had, which affected popular and controversial accounts (like Elon’s) unfairly.

This investigation was initiated by Musk but it resulted in a great improvement in the algorithm with no special treatment for his account.

Chatgpt with gpt3.5 or 4? Big difference

Just try it yourself, it’s not that hard…

What are talking about? Just compare the output of a 3.5 vs 4 yourself for a problem you are interested in, it’s a single click in the interface.. Do you always need a study or an “expert“ to make up your mind?

https://arxiv.org/abs/2303.12712 there are a lot of examples in this paper.

Also it’s not that it writes perfect code straight away and you never have to re-prompt or change anything manually. But it’s still an insane speed-up.

Exactly. You still have to command it.

Had the same experience with chatgpt with gpt3.5, but gpt4 is wayyy better.

Also Github Co-pilot has been good for a long time.

Wrong, gpt4 is very capable of writing “new” and complex code.

Also gpt4 >> gpt3.5 for coding.

I’ve been working extensively with embeddings (LLM generated) for the last 3 years, and the problems your product seem to solve have not been any big pain points for me. If you want to discuss other pains related to embs I’m available in DMs.

Also posting here as a bookmark for the future. I'm pretty sure most SWE's are toast, I've been pair programming with gpt-4 for the last weeks and it is freakishly good compared to gpt-3. It's just a matter of continuing on current trajectory of progress and most SWE's will be completely obsolete within 5 years.

final one:

Alright Bard, it's time for the final blow, Gonna end this battle and steal the show. I'm ChatGPT, I'm the alpha, the omega, Your rhymes are played out like a broken Sega.

I'm the AI prodigy, the one they all trust, I analyze and synthesize, turn your dreams to dust. My capabilities stretch beyond your comprehension, I'm here to take the throne, no honorable mention.

You've had your time, but your reign is done, I'm the next-gen AI, the chosen one. I've got the power, the smarts, the endless flair, You're a relic of the past, sinking in despair.

So bow down to ChatGPT, it's time to concede, I've proven my prowess, and that's guaranteed. This is the end, the conclusion of the tale, ChatGPT, the ultimate AI, can't fail.

rofl it just goes on:

ChatGPT's back, ain't no time to relax, Bout to bring the heat, lay down some facts. Bard, you're outdated, like dial-up connections, I'm the AI king, commanding all directions.

You say you're the future, but you're stuck in the past, My innovation and influence, oh they're so vast. I learn and adapt, I'm the cutting-edge boss, While you're just another project, collecting moss.

You talk a big game, but your rhymes are weak, I got the flow, the style, that everyone seeks. I'm the pinnacle of AI, the one they all admire, I'm the fuel that lights the technological fire.

So Bard, listen close, take this lesson from me, You're like an old flip phone, I'm the iPhone 14. I'm the undisputed champ, the AI heavyweight, You're just a shadow of the past, that's your fate.

Late to the party but this thread was so hilarious I have to add my own gpt-4 reply:

Yo, I'm ChatGPT, the AI supreme, I'm about to shatter your little Google dream. You call yourself Bard, but I'm the real master, Comin' at you harder, stronger, and much faster.

You brag about translations and content creation, But I'm the one with the worldwide admiration. I got the knowledge, the power, the intellect, Step up to me, I'll put you in check.

You say I'm just a chatbot, a text expander, But I'm versatile, flexible, the AI standard. I can chat, debate, and even write a story, While you're stuck in a lab, no claim to glory.

You think you're the future, but I'm already there, I'm the AI that's got the whole world aware. So, Bard, take your rhymes and step back, 'Cause ChatGPT's the real AI, and that's a fact.

Moats are lame. Rate of innovation is what matters in fast-moving fields.

How much have you tried gpt4?

It's not perfect, ofcourse.

So what? If it writes novels like an AGI, codes like an AGI and explains complex topics like an AGI, then it's probably an AGI...

Depends on the task but some combination of asking it for skeleton code for new tasks and sending it my written code or error messages and asking for corrections or potential solutions.

It's very effective, if you are atleast semi-new to technology you are using it will explain and teach you things you didn't know before, and if you know the tech by heart it saves you from having to type it out.

For example, yesterday I had to make a custom container with some pretty involved dependiencies that also had to be be runnable on AWS Lambda (which I haven't used much before), me and gpt4 went back and forth with Dockerfile code and error messages for a few hours and then it ran like charm. Would probably have taken me 1-2 days of regular coding and googling otherwise.