Hacker Newsnew | past | comments | ask | show | jobs | submit | patrick-fitz's commentslogin

I'd be interested to see how it performs on https://www.swebench.com/

Using SWE-agent + Yi-Coder-9B-Chat.


Sorry, but the landing page is a bit dull. I think you need a video showing a live demo of the product.


That's a good point. We built repaint.com in Repaint and we don't currently have videos in our editor. We will likely add a demo video on the site once we can.


I think they should be a distinction between open-source and open-weight LLM's.


I like this terminology, I'm going to start using it.


The problem is, all the Open Weight models are already calling themselves Open Source, so a new name that disambiguates existing names should be chosen.


Thanks for the link, this does make it more approachable!


When clicking on a .org domain, the website redirects me to a .org.uk domain on Namecheap.


Same with com.de instead of com


I think Meta.ai needs to add a disclaimer below the chat input box similar to ChatGPT:

"ChatGPT can make mistakes. Consider checking important information."


The disclaimer really should be much tougher: "Every LLM consistently makes mistakes. The mistakes will often look very plausible. NEVER TRUST ANY LLM OUTPUT."


> NEVER TRUST ANY LLM OUTPUT

that doesn't sound like a helpful attitude. everything you read might be wrong, llm or not - it's just a numbers game. with gpt3 i'll trust the output a certain amount. it's still useful for some tasks but not that many. gpt4 i'll trust the output more


LLMs are impressively good at confidently stating false information as fact though. They use niche terminology from a field, cite made-up sources and events, and speak to the layman as convincingly knowledgable on a subject as anyone else who's actually an expert.

People are trusting LLM output more than they should be. And search engines that people have historically used to find information are trying to replace results with LLM output. Most people don't know how LLMs work, or how their search engine is getting the information it's telling them. Many people won't be able to tell the difference between the scraped web snippets Google has shown for years versus a response from an LLM.

It's not even an occasional bug with LLMs, it's practically the rule. They don't know anything so they'll never say "I don't know" or give any indication of when something they say is trustworthy or not.


at least the llm (for now) doesn't have an agenda

the top result on google is literally just the result of how hard someone worked on their seo. they might not "hallucinate", but a company can certainly use strong seo skills to push whatever product/opinion best suits them.


But it’s correct. Without independent verification, you can never, ever trust anything that the magic robot tells you. Of course this may not matter so much for very low-stakes applications, but it is still the case.


They have now added a disclaimer: "Messages are generated by AI and may be inaccurate or inappropriate."


A disclaimer isn't about solving a problem; it's about absolving responsibility.


> Consider checking important information.

How do you check important information, when it's all wrong? And when companies are pushing LLMs as where you go to check this information?


> How do you check important information, when it's all wrong?

It's not very helpful to say that it's all wrong. It isn't the case, otherwise there would not be any issue. Whether the right answer is produced by reasoning or a statistical model does not take the answer not-right.


But basically nothing in these answers was "right". It was merely plausible. "Important information" or not, what is the point of this garbage?


Good idea, that'll make sure people don't share hallucinations as facts /s


Nice project! I find it can be hard to think of a idea that is well suited to use AI. Using embeddings for search is definitely a good option to start with.


I made a reverse image search when I learned about embeddings. It's pretty fun to work with images https://medium.com/@christophe.smet1/finding-dirty-xtc-with-...


https://twitter.com/karpathy/status/1777427944971083809

> And once this is a in a bit more stable state: videos on building this in more detail and from scratch.

Looking forward to watching the videos.


I love his videos. They are dense, but I get a lot out of them.


+100 thank you karpathy!


Out of curiosity, looking at the cheapest price for a H100 that I could find online.

Lambda Reserved Cloud [1] starts at $1.89 per H100 per hour.

It could be possible to get the cost down to a lower amount:

$1.89 * 96GPUs * 24hours * 14days = ~$61k

1 - https://lambdalabs.com/deep-learning/servers/hyperplane


This is the price of training if nothing fails.


It also depends on the interconnect speed. If you don't have fast enough interconnect between the machines, you won't get linear speedup with N gpus.


> (The New York Times has sued Microsoft and its partner OpenAI on claims of copyright infringement involving artificial intelligence systems that generate text.)

It's strange to see this included randomly in the middle of the article.


Conflict of interest disclaimer. It's pretty standard in journalism.


It's run-of-the-mill disclosure.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: