ebalit

Generative methods have usually been considered unsupervised.

You're right that conditional generation start to blur the lines though.

I've built exactly this as an experiment (code interpreter with pyodide sandbox). Should probably tidy it up to open-source it.

Are you sure it was a chat bot? I believe Apple customer service chat wasn't run by bots last time I needed it.

It really depends on the release. It's not the case for the base Mistral 7B model. But many finetuned models are released with some information about the dataset used.

Both would be even better but it would be much smaller progress for the open source community in comparison to the release of the weights.

Mistral 7B [1] and many models stemming from it are released under permissive Apache license.

Some might argue that a "pure" open-source would require the dataset and the training "recipe" as it would be needed to reproduce the training, but it would be so expensive that most people wouldn't be able to do much with it.

IMO, a release with open weights without the "source" is much better than the opposite, a release with open source and no trained weights.

And it's not like there was no progress on the open dataset front: - Together just released RedPajama V2, with enough tokens to train a very sizeable base model. - Tsinghua released UltraFeedback which allowed more people to align models using RLHF methods (like the Zephyr models from Hugging Face) - and many many others

[1] https://mistral.ai/news/announcing-mistral-7b/ [2] https://github.com/togethercomputer/RedPajama-Data

I know the answer is a bit tongue in cheek, but that would be very valuable and cool, just in case someone from Cursor or Refact is reading this thread ^^

That's more expensive than Paris. Well it depends the type of restaurant of course. Would you have some examples in Bucharest?

GPT4 can be used to describe your tone by analyzing a few of your emails.

Adding that to your prompt should help to get something closer to what you would write, and you can iterate on this description if needed.

Source: I built a chatgpt ui that make it easy to create custom prompt template because I was tired of the copy-paste and/or back and forth to get the tone right.

I saw once the proposal that revenues from a carbon tax should be redistributed equally.

It seems like a practical implementation of a bonus-malus system.

Yes that was an interesting part.

Here in France it seems like the apps say the delivery should be upstairs if the customer asked for that, but drivers have the same issue with going upstairs as in Brazil.

And I can see why if they get paid the same.

My guess is, the apps know the issue, but explicit "street-level" deliveries don't sell as well.

It seems like "upstairs" delivery should be a paid option and be explicit in the ordering and delivery apps.

Great insights. It should be the top comment IMO. That's the kind of field-tested perspectives that give HN its value.

I wrote about that "platform play" a few month ago with a different take [0].

They could have made a "Connect with OpenAI" scheme so that developers can use the user OpenAI API directly.

That way developers could focus on the UX, they could focus on the LLM, and users would get a centralized discovery / billing for their LLM based tools.

I'm probably missing something that would have prevented that strategy but I think that would have been much stronger than the plugins.

And I'm really not sure that it would still be possible 5 month later.

[0] https://www.linkedin.com/posts/etienne-balit_ceo-at-open-aic...

A "bunch of words" has been the standard for the dissemination of scientific results for decades. Worse, those results are often locked behind a paywall...

Publishing models is definitely better than only a paper, but I think you're being unnecessarily harsh in that case.

We would all prefer it if OpenAI released a proper paper about GPT4, for instance, even if they did not release the complete model alongside it.

What kind of payment method are you looking for?

It looks like a description of Speculative Sampling. There's a recent paper from DeepMind about this in the context of LLM [0], although it's not a completely new idea of course.

The potential for speedup according to their paper is closer to 2x than 10x however.

0: https://arxiv.org/abs/2302.01318

It's a side effect of the way the text input is represented before being used by the model. It doesn't get the text as a sequence of chars but as a sequence of tokens.

This paper [1] shows that giving character-level awareness to the model can improve the "visual spelling".

1: https://arxiv.org/abs/2212.10562

Reactivated [1] looks similar in spirit, but using React on the frontend. I haven't tested it yet but the idea is very interesting.

[1] https://www.reactivated.io/

Not only new users. I was logged off after reloading the page.

It's not AI but it's definitely a tool to build AI systems.

Similarly, GPUs or CUDA aren't AI in themself but are fundamental in the day to day work of AI practitioners.

It might be very interesting to write the "no value" comments automatically. It can help us focus on and emphasize the part that we think is missing from the autogenerated ones.

This is the first time I read this and I'm very surprised that it is the case. Would you have a source I could read to learn more about this.

Given the wavy logo, I guess it reads as "Seven Seas"?

2.5x seems really high, but 2x is usual for software engineer salaries and up.

For example, an average software engineer can receive a gross salary of 50 k€ which translate to 34 k€ after social contributions and income taxes (in a worst case scenario) and a total cost of 71 k€ for the company [0].

Using this simulator, I can reach 2.5x by putting in a gross salary of 250 k€. Of course it's mostly because of the high income tax in this case.

There are also strong incentives for startups, most notably reduced social contributions during their first 8 years [1] and tax credit for a third of research expenses [2].

Admittedly, this is probably more complex than needed but I think it is still way less expensive than hiring in the US.

0: https://mycompanyinfrance.fr/calculators/salary?salaire-brut...

1: https://www.service-public.fr/professionnels-entreprises/vos...

2: https://www.service-public.fr/professionnels-entreprises/vos...

Apple GUI frameworks enable them to rely heavily on the GPU, and the smooth and slick GUI are definitely a big part of their value proposition. While I agree that the business logic of most apps aren't using the GPU for compute, I think most app will have a rendering side where it is important.

Pascal is also a first name in France even though in the case of the programming language, it is named after Blaise Pascal last name.

Not commenting on Moore's law, but I don't see how the last part is true when GPU have become a main component of personal computing. They are in fact becoming more and more important with the need for AI inference and always more demanding rendering tasks (both in resolution and frame rate).

Saying they didn't lie but that they misled the public is really nitpicky. They betrayed the trust that people had in their opinion. Maybe it was worth it to manipulate people "for their own good" but I doubt it.

It was indeed what I was referring to.