How can they speak of Safety when they are based partly in a colonialist settler entity that is committing a genocide and wanting to exterminate the indigenous population to make room for the Greater Zionist State.
I don't do business with Israeli companies while Israel is engaged in mass Extermination of a human population they treat as dogs.
During training, it's more efficient than full finetuning because you only update a fraction of the parameters via backprop.
During inference, it can ...
1) ... be theoretically a tad slower if you add the LoRA values dynamically during the forward pass (however, this is also an advantage if you want to keep a separate small weight set per customer, for example; you run only one large base model and can apply the different LoRA weights per customer on the fly)
2) ... have the exact same performance as the base model if you merge the LoRA weights back with the base model.
It is legal. Fair use. People have been doing it for ages. Almost every article you've ever read has some fair use of another article, book or news item, etc.
The Tolkien estate should get busy suing all the fantasy writers, comic artists, game developers and board and card game companies. Lots of cash there.
They have done some of that actually. Tolkien will be public domain in the nations that are at aithors death+50 in a few days. Sadly, it will be a much longer wait in mine and many others.
Browser vendors are (or should be) managing the abstractions for their own needs, with developer needs expected to be met by framework/library developers.
Who says web components are meant for use directly by the developer? Maybe they're primarily meant for the browser developers (those who build browser features), not for use directly by web app developers.
No. Browser vendors implement specs, defined by the W3C. To quote the W3C‘s tagline:
> The World Wide Web Consortium (W3C) develops standards and guidelines to help everyone build a web based on the principles of accessibility, internationalization, privacy and security.
Web Components are a standard to be used by web devs.
FYI: the last time W3C, Inc. endorsed a WHATWG HTML snapshot as a recommendation was in 2021 ([1]), for the WHATWG snapshot published January, 2020 whereas the 2021 and 2022 snapshots were rejected, with no new review process having started since.
Reasons for rejection include
a reporting API seen as privacy-invading and related disagreement over whether W3C's HTML WG could augment/redact WHATWG spec text like they used to be doing until 2017 when the previous HTML recommendation was published, or has to go through WHATWG process for any change according to the W3C/WHATWG "memorandum of understanding" which thus hasn't resulted in common understanding after all ;) Another reason was objection against the so-called HTML5 outlining algorithm, which Steve Faulkner actually has gone to great lenghts removing in WHATWG HTML upstream (cf. [2] for details).
Unfortunately, the removal also brought incompatible change to HTML (the content model of hgroup, among other incompatibilities), rendering existing content invalid, which WHATWG set out not to be doing but which they lack the methodology of preventing and for which the spec derived from Ian Hickson's work frankly lacks formal qualities to support. Even more unfortunate is that this change has already spilled to derived standards such as EPUB3 which hence makes existing EPUB3 content using compound headings going back to 2011 invalid, and EPUB3 writers lacking a tool for actually verifying what readers can support (epubcheck was blindly updated without consideration for the installed base). Technically, Review Draft January 2022 and newer should then already be called HTML 6. Since nobody gives a rat's ass (including W3C, Inc.'s dormant HTML WG) anyway, and gross misconceptions about HTML specs prevail, like in your post, I'm not sure whether we should call it a day with WHATWG/W3C's HTML specs already.
And who defines the W3C specs? Browser vendors! And if they don't agree with the non browser vendors within W3C, they create another standard org (WHATWG).
Kind of the point. WHATWG is descriptive not prescriptive. It describes things as they are. W3C is a standards body, sometimes they use the existing solutions and standardize them.
This is why I abandoned my career writing JavaScript. It’s an industry of people hopelessly in need of frameworks and tools to do their job for them because they cannot program.
I think that's unfair. There are a lot of Javascript programmers that are perfectly competent.
Unfortunately, the trash "become a programmer in three days" bootcamps and scam courses all target web development, flooding the market with people who were taught one or two tricks and told they're the cream of the crop and should definitely not ask for their money back if nobody wants to hire them.
It's the same problem PHP and Python have suffered from: when your programming language and API is accessible and easy to use, you'll attract a lot of beginners and people who skipped the hard parts.
For some reason, the frontend world seems intent on reinventing itself every five years or so. The backend world works in cycles of 10 to 20 years, but it's going through the same motions. Everything became C, then C++ and Delphi came along, then everything became Java and DotNet, mow everything is becoming Go and Rust, and every iteration brings about new design concepts and paradigms.
What business doesn't use some kind of framework, even for the backend? I've never heard of a successful business building everything from scratch, usually you'll have something like Qt for GUI construction, or Django/Ktor/ASP.NET for a web server.
I believe Wikipedia uses mostly frameworkless Javascript, but, as you might expect from any sufficiently sized project, they have come up with their own frameworks instead.
Well, if it's so useless why is it on the HN front page? Are there "PR" companies behind promoting items to the HN front page? I'm sure there are because sometimes an article like this comes up at #3 and everyone says it's got no substance, clickbait, etc
I'll take the "PR Company" thing as a compliment. But no, afraid not. I'm one of the two founders and I wrote this one - sorry if it came across as click-baity, but I just wanted to outline the basic, long term idea that Hivekit is going for.
"Four years ago, Altman’s mentor, Y Combinator founder Paul Graham, flew from the United Kingdom to San Francisco to give his protégé the boot, according to three people familiar with the incident, which has not been previously reported."
I guess he had a change of heart about Sam because ... ?
Sam didn't create the breakthroughs behind the current GPT.
He did not create the breakthroughs behind the next GPT.
None of the people that may follow have the same handle on the tech as Ilya. I mean they built up Ilya's image in our mind so much, that he's one of a kind genius (or maybe Musk did that) and now we are to believe that his genius doesn't matter and that Microsoft already knows how to create AGI and that OpenAI is no longer relevant?
You are assuming he wouldn't steal t from OpenAI. He could have a low level employee steal it, and manage to keep it a secret until AGI is born then he takes over the world.
This is a pretty wild comment. That's a very safe assumption and no low level employee will do Sams bidding in an illegal enterprise. And keeping it a secret isn't going to work either and whether or not AGI is 'born' (who will bear it) is an open question to which I hope the answer is 'not for a while'. Because we haven't even figured out how to get humans to cooperate which I think should be a prerequisite.
> no low level employee will do Sams bidding in an illegal enterprise
Many people have betrayed their country to foreign governments in exchange for mere thousands of dollars. It is never safe to rule out the willingness of employees to engage in corporate espionage, even in exchange for truly pitiful rewards. It would be a stupid idea, but that doesn't mean it won't happen.
What did Sam Altman hide from his board that caused his firing as CEO of OpenAI?
1) That LLMs cannot generalize outside of _patterns_ they pick up during training? (as shown by a recent paper from Google, and as many of us know from our work testing LLMs and working around their short comings)
2) That every time you train a new model, with potentially very high expense, you have no idea what you're going to get. Generally better but also potentially bigger reliability challenges. LLMs are fundamentally unreliable and not stable in any kind of use case besides chat apps, especially when they keep tweaking and updating the model and deprecating old ones. No one can build on shifting sands.
3) The GPT4-Turbo regressed on code generation performance and the 128K window is only usable up to 16K (but for me in use cases more compicated than Q&A over docs, I found that 1.2K is max usable window. That's 100X than he advertised.
4) That he priced GPT4-V at a massive loss to crush the competition
5) That he rushed the GPT Builder product, causing massive drain on resources dedicated to existing customers, and having to halt sign ups, even with a $29B investment riding on the grwoth of the user base. Any one of the above or none of the above.
No one knows... but the board.. .and Microsoft who has 49% control of the board.
Repo has been deleted? I get a 404. I did see it earlier on.
I fed the equation image (screenshot at the right frame from their gif then cropped) into ChatGPT (GPT4-V) and it correctly deciphered the equation and gave the correct LaText code.
I don't do business with Israeli companies while Israel is engaged in mass Extermination of a human population they treat as dogs.