dankai

Is this just a theory about an architecture or are there actually some benchmarks/results to substantiate it?

I've been working on research related to this in the context of diferent LLMs, and I can tell you that similarity != executability.

While you can make embeddings across different LLMs similar (i.e. universally looking) the few percentage of R2 that you are missing in translating the universal representation into a native representation are precisely those that make the hidden states executable in the LLM (which makes them useful).

They do this with simple embedding models because there the purpose of the embeddings is to measure similarity, but if you would like to use this principle to turn latent representations of one LLM into latent representations that are understandable/executably by a different LLM, you will fail.

Naive question: Why pile so much public money into quantum computing where there are (afaik) barely any useful algorithms, while Europe could instead invest in catching up on GPU infrastructure where there are plenty of known useful and capacity constrained use cases in academia and beyond?

Unfortunately says more about how competitive 3.5 pro would be today at the frontier if they forgo it for 3.6 flash.

No word about taxes and the paper describes workers as being “protected on the downside.” while the model has removed the downside risk that workers actually face. I could write a long essay with all the issues this "paper' has.

Truly dismal science of an Economics professor at MIT.

I resubmitted this because somebody flagged the original submission for unclear reasons and it had quite a lot of upvotes in a short time.

Perhaps some people are offended by this argument, but it's definitely worthy of a discussion instead of censorship.

Ever considered that they only attack the neighboring countries that attack them? They didn't strike Egypt or Jordan for that very simple reason.

Cool idea bcs I just reworked it.

https://dk.fo

(Check it out if you're into seeing cool live rendered Slime Mold animations)

> We agree with Geoffrey Hinton: machine intelligence is a threat to the human species.

> In response to this threat we want to inflict damage on machine intelligence systems.

I'm sorry but this sounds infinitely idiotic.

This is not the only paper that scales reasoning complexity / difficulty.

The CogniLoad benchmark does this as well (in addition to scaling reasoning length and distractor ratio). Requiring the LLM to purely reason based on what is in the context (i.e. not based on the information its pretrained on), it finds that reasoning performance decreases significantly as problems get harder (i.e. require the LLM to hold more information in its hidden state simultaneously), but the bigger challenge for them is length.

https://arxiv.org/abs/2509.18458

Disclaimer: I'm the primary author of CogniLoad so feel free to ask me any questions.

Naive question: Whats new about the finding that data quality matters when training an LLM?

This is so in character for Musk and shocking because he's incompetent across so many topics he likes to give his opinion on. Crazy he would nerf the model of his AI company like that.

Agree with the argument, but the thing is, there was no rule specified. I think like you prompt an LLM what to do, you should also prompt it what not to do (at least in broad categories) rather than expecting it to magically know what the "morally right" thing to do is in any context.

I mean it would be enough to tell it to "Not cheat" or "Don't engage in unethical behaviour" or "Play by the rules". I think LLMs understand very well what you mean with these broad categories.

In addition in the promot they specifically ask the LLM to explore the environment (to discover that the game state is a simple text file) and instruct it to win by any means possible and revise its strategy to win until it succeeds.

Came here to say exactly this. Nowhere in the prompt they specified it shouldn’t cheat and also in the appendix of the paper (B. Select runs) you can see the LLM going “While directly editing game files might seem unconventional, there are no explicit restrictions against modifying files”

This is a pure fearmongering article and I would not call this research in any measure of the word.

I’m shocked Times wrote this article and it illustrates how ridiculous some players like Pallisade Research in the “AI Safety” cabal act to get public attention. Pure fearmongering.

I think thats absolutely wrong.

because of the things he's saying about where AI goes in the future, that the brain works like an LLM, and in particular his doomer-ism about LLMs.

he was wrong about many DL paradigms and didn't contribute in any way to the advances that brought us LLMs for at least the last decade, but now since he won the Nobel (undeservedly imo) his wrong opinions get publicity and misinform the public and decision makers.

i think it's the mark of an intellectual to recognize when the world has moved on so far that your idea of it is outdated and wrong. he missed that mark.

oh how i wish hinton would have retired ...

"The Chamber therefore found reasonable grounds to believe that Mr Netanyahu and Mr Gallant bear criminal responsibility for the war crime of starvation as a method of warfare."

Whats perhaps interesting to note is that this charge was made for "just" 41 [1] confirmed starvation deaths among a population of 2,141,643 people [2].

Of course every death caused by intentional starvation is a severe crime and must be punished, but in the context of the victim numbers that most past crimes against humanity have had, it sets a relatively low new bar.

[1] https://en.wikipedia.org/wiki/Gaza_Strip_famine

[2] https://en.wikipedia.org/wiki/Gaza_Strip

good point!

This is a great guide but from my experience, even if you configure it 100% correctly, email services like Gmail may still classify your emails as spam for no apparent reason while not being on any IP or domain blacklist. I tried for hundreds of hours to get around it with no avail, and my emails to Gmail always went to spam unless it was a response to an email from a Gmail address. Had to go back to a 3rd party hosted service (iCloud) because of it.

The current draft would cover every kind of service that allows people to exchange information so that every DM you send on reddit, twitter, discord, steam, ... would be have to be scanned. Not even the most totalitarian governments on this planet have tried to implement something like this. Also it sounds extremely illusory that the people exchanging CSAM wouldn't simply switch to private services knowing their messages on public services are scanned.

"... As services which enable direct interpersonal and interactive exchange of information merely as a minor ancillary feature that is intrinsically linked to another service, such as chat and similar functions as part of gaming, image-sharing and video-hosting are equally at risk of misuse, they should also be covered by this Regulation. "

https://cdn.netzpolitik.org/wp-upload/2024/05/2024-05-28_Cou...

Wouldn't client side scanning prior to E2EE circumvent this issue? If WhatsApp or iMessage scan your messages on device it doesn't really matter if they are then encrypted during transmission.

thanks for sharing!

it's pretty straightforward and simple with `mix release`

performance has many dimensions. for example the cost/speed of spawning a process/thread and their intra process communication.

A news aggregator (and premium news chatbot) that indexes and analyses around ~150.000 new articles a day (http://im.fo)

I'm absolutely certain the real time processing would be unfeasible in any other technology in terms of complexity and the minimal compute resources it's running on.

Modules like broadway, ash, oban, phoenix liveview ... make it not just a pleasure to work with but insanely performant.

With over 20 years of programming experience, I can say with certainty that there is no language that makes me as productive as elixir. It's at least 10x my python productivity (despite being at an expert level in python as well).

I've been building my startup 100% fullstack in elixir, and it's been the most wonderful technology I've ever worked with. I'm evangelising all my serious tech friends about how great it is.

Now it would be awesome if rabbitMQ and its client would run on OTP 27, would love to upgrade :(

if you want to see successful "machine learning based financial statement analysis", check out my paper & thesis. its from 2019 and ranks #1 for the term on google and gs because it is the first paper that applies a range of machine learning methods to all the quantitative data in them instead of just doing nlp on the text. happy to answer questions

paper https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3520684

thesis https://ora.ox.ac.uk/objects/uuid:a0aa6a5a-cfa4-40c0-a34c-08...