aubanel

I love the example with Russian elections! It's impressive that they allow themselves to be that low-effort in their fake results.

Did learning about the brain being made of meat also destroy your awe of human intelligence?

I feel like being able to look inside the clockwork should not distract us from being amazed at how wonderful a system can be.

The Rafale is cheaper than the F35, by a wide margin.

> My New Year's Resolution in 2019 was to learn the piano. Key to success is building mental representations of music and the keyboard. Increasingly, I do this by playing pieces on my mental piano

As a correct pianist (lots of classical study, and sometimes I am paid to play in bars), this seems very weird to me. I don't see how doing this could be any useful for learning, it seems like a waste of effort.

Systematically calling russian people Orcs is the worst dehumanising things to do IMO.

I have nothing to do with that conflict but I don't like how confident some people are that they're on the side of the Good, thus that they could do whatever to their opponents. (Cf Kasparov who wanted to nuke Russia at the beginning of the war)

I prefer my theory, it's funnier although less probable, I agree

I love how they show the failure cases: compare that with Gemini 1.5 pro's technical paper that carefully avoids any test where it does not seem like a 100% perf! I think confronting your failures a condition for success, and Google seems much too self-indulgent here.

> They are pretty clear that 1.5 Pro is better than GPT-4 in general, and therefore we have a new LLM-as-judge leader, which is pretty interesting

I fully disagree, they compare Gemini 1.5 Pro and GPT4 only on context length, not on other tasks where they compare it only to other Gemini which is a strange self-own.

I'm convinced that if they do not show the results against GPT4/Claude, it is because they do not look good.

For reference, here is the technical report: https://storage.googleapis.com/deepmind-media/gemini/gemini_...

New theory: Karpathy was short on money, so he waited exactly 1 year to vest 25% of his options.

I think the end goal of self driving cars is simply to let people do something else than drive. Safety is only a condition for this to happen.

HF paper page: https://huggingface.co/papers/2401.09417

"Brains are not required to think or solve problems" is a perfect intro to a Leetcode job interview, for a candidate that already grokked 300 of these problems.

Thank you, great comment!

Anyone has insights on which of these networks really have gained a strong, stable user base? I hear a lot on mastodon for instance, is it really becoming bigger or staying very niche?

I often hear the argument "the destroyed jobs will be replaced easily, as has been the case in all technological revolutions", but I think it's deeply flawed:

Up to know, all technological revolutions have occurred in an exponentially growing economy. That's why Schumpeter's "creative destruction" was not a problem: it was compensated by the opening of new areas for growth, mostly unlocked by consuming more resources.

But now we are hitting the limits of the planet, and amongst others the limit of extraction rhythm for most our materials. So the exponential growth can't continue. (apart from the growth from technological progress, which will be much lower)

That's where the argument doesn't hold: we do not have a growing supply of work anymore, so the increased productivity of some workers will just prevent the others from working.

Look at this: https://huggingface.co/spaces/multimodalart/Ip-Adapter-FaceI...

From a few crappy photos of your face, you can generate photos of you in nearly every setting: so I think all the "studio" part of a commercial photographer's job is toast.

Don't forget Rule n°1 : no pre-orders!

A datapoint I find impressive on the debate "What should we privatize": UK Rail privatisation seems to have worked really well, as shown in this graph: https://upload.wikimedia.org/wikipedia/commons/9/9f/GBR_rail...

If you have the most powerful model by far like they have, it's a radical adantage over competitors: your model can perform automatizations that they cannot do.

OpenAI's lead is phenomenal: remember that they managed to put GPT4 on top of the leaderboard for one year and counting, while all others (except maybe Anthropic) struggle to have their model in the top3 for more than a few months.

This "Journalist X LLMs" video gives the same corporate, polished vibe as the Gemini announcement... Google's AI communication really feels like any "snake oil consulting group" at this point. I really hope they get their stuff together and start putting engineers & researchers in charge again.

A website allowing to browse chords for variety songs (mostly french songs): le-chansonnier.com

Mistral sure does not bother too much with explanations, but this style gives me much more confidence in the product than Google's polished, corporate, soulless announcement of Gemini!

Other people with similar experiences at FAANG? About Google, I still cannot yet grasp if the company is just starting to slow down or already declining steeply.

Indeed it seems like replacing "hallucinate" with "confabulate" would improve understanding of the phenomenon in english, but it could cause a problem in Romance languages.

Indeed, the Latin verb "confabulor has the meaning of "discuss, converse" [0], not specifically "inventing things", and Romance languages already have words derived in sound and meaning from this classic Latin sense. For instance French already has "confabuler" meaning "speak familiarly with someone" [1]. Thus in Romance languages, this proposed usage of "confabulate" imported from the psychiatric word could be understood poorly due to preexisting similar-sounding words that do not include the notion of "making things up".

Just proposing thoughts here, not advocating for a final decision. Maybe it's still good to adopt this wording!

[0] : https://en.m.wiktionary.org/wiki/confabulor#Latin [1] : https://www.dictionnaire-academie.fr/article/A9C3479

This page has an original way to force the user to accept cookies : present the cookie banner in a way that blocks the content. If the user clicks "Refuse optional cookies", just reload the page and present them the same choice again. That is illegal in the EU now.

Congrats to them, their huge series A (USD113M) does not seem oversized anymore!

The father of a friend worked at a high position at Canon. For a specific enterprise command, the client company wanted printers with a lifetime 20,000 copies instead of the base 10,000. Canon went "no problem, we'll see with our engineers" and actually they only had to remove a small device which was basically a print counter and would artificially block the ink input at approx 10,000 copies. So nothing to do with good engineering, only bad business practices.

These precepts are the "Three sieves", often wrongly attributed to Socrates. Here [0] is an interesting discussion on their origin. For instance, these three sieves are regularly referenced in Europe since the XIXth century, but can already be found much earlier in the Indian Manusmriti.

[0]: https://skeptics.stackexchange.com/questions/39843/are-socra...

The examples cited here to illustrate similarities between chess and war are just that : similarities. Juste as one could see similarities with a basketball game. The differences between chess and war are so numerous (chess has no fog of war, no command issue, no logistics, no morale...) that I'm starting to dismiss every "generals can learn from the game of chess" piece as bullshit.