HNHacker News
TopNewBestAskShowJobs

atemerev

7,477 karma · joined September 17, 2013

Founder at Virial, ex-EPFL
submissionscomments
atemerev··on The LLMentalist Effect (2023)
What is "intelligence" then. I think the universalism of LLMs is now evident enough that they can be considered "intelligent", even in slightly different realization compared to humans.

As of "great damage", I doubt it. They do not have self-preservation instinct (all "worrying" experiments are the attempts to initiate something resembling self-preservation from human initiative). The driving part is external - the query loop can always be turned off. So yes, a dangerous tool that can be exploited by humans (including governments, especially governments - which is why I am skeptical to government regulation proposals, particularly looking at what passes as governments in this era). But there are no inherent dangers from their own agency, as there is none.

atemerev··on The LLMentalist Effect (2023)
All of this is accurate (and useful). But yeah, psychics do not prove theorems from frontier math and build working complex software.
atemerev··on South Africa is at risk of becoming a mafia state
This is not feudalism. Feudalism requires people to exploit and land domains to control. "Tech feudalism" requires neither.
atemerev··on South Africa is at risk of becoming a mafia state
As someone who was born in Russia and lived there in 1990s, I have quite a good idea of this arrangement. This was not mafia state yet. The mafia state is the modern Russia, where former organized crime groups became fully integrated into the government. Technically, the state retains the monopoly on violence. Practically, violence comes from influence of competing groups within the state. And yes, US is moving towards the same regime type. This is clearly visible to me, from my Russian experience, but I don't understand how to convey this to Americans.
atemerev··on South Africa is at risk of becoming a mafia state
United States of America are at risk of becoming mafia state. Very obvious and very significant, with fewer and fewer possibilities of reversal remaining each day.
atemerev··on Xiaomi Mimo 2.6 live post-training dashboard
No crying in the copyright casino.
atemerev··on Xiaomi Mimo 2.6 live post-training dashboard
Ah, one Donald Trump, a famous neoliberal.
atemerev··on XCancel service is suspended until further notice
Well I have heard that some people managed to sue some corporations and actually win some cases before.
atemerev··on XCancel service is suspended until further notice
If a society operates under a rule like this, it is no better than Russia or any other tyranny where law is for me but not for thee. This is not how it should work in a supposedly free and lawful country.
atemerev··on Flock used >100 times to track veteran who recorded traffic stop
Still not enough. What you suggest does not prevent the government of possessing the data. Technical means of collection needs to be preemptively banned.
atemerev··on Flock used >100 times to track veteran who recorded traffic stop
Not enough. Preemptively passing laws that forbid this kind of mass surveillance. Like in Switzerland and other nice places.
atemerev··on GPT-6 Astra
Damn I hate this benchmark. SVG authoring from head without visual reference is so wrongly posed.
atemerev··on GPT-6 Astra
They simply refuse my applications to slightly less restricted models without any explanations. And the current ones refuse automatically to work with me on my papers as soon as they see the word "epidemiology".

I am a researcher in a Swiss university btw.

atemerev··on Muse Spark 1.3
Ok. This requires the notion of "intrinsic value", which I believe does not exist (all value is subjective), yet is a foundation of all Marxist theory.
atemerev··on Muse Spark 1.3
Why "terminate", Marxist philosophy is a legitimate topic, deserving to be studied. Like a rich sci-fi lore or a history of Tarot magic. Deep, fascinating, and wrong.
atemerev··on Gemini 3.8 Flash and 3.8 Flash Cyber
"Uncensored" means "weights modified to remove refusals". Abliterated. Providers do not serve such models, at least not frontier-grade. You have to run the weights yourself. For Kimi K3, this is about $60/hour for hardware rental. But you can have about 100 sessions simultaneously.

And yes, Fable 5.1 has the same refusal rate, and significantly nerfed reasoning.

atemerev··on Gemini 3.8 Flash and 3.8 Flash Cyber
A/S/L?
atemerev··on Gemini 3.8 Flash and 3.8 Flash Cyber
Everyone is censoring models now with anything remotely resembling cyber or bio. I already have problems with my research in mathematical epidemiology because of that - both Sol and Fable simply refuse. They keep pushing people towards Chinese models that can be decensored.
atemerev··on A 3D fruit fly on macOS desktop powered by the real FlyWire connectome
A human brain is about 100T parameters (if you count every synapse) on fp4 quantization (the amount of state information available in every synapse). The architecture is quite different from modern LLMs and currently not well enough understood, but _theoretically_ a modern GB300-NVL72 cluster is nearly computationally sufficient to simulate a human brain (with some optimizations - almost certainly is).

For state equivalence, we don't need to simulate every chemical reaction in every brain cell.

atemerev··on A 3D fruit fly on macOS desktop powered by the real FlyWire connectome
Depends on the simulation level. A lot of things can be abstracted away, we don't need to simulate everything down to atomic level.
atemerev··on "Sabotage": Experts, lawmakers blast RFK Jr. for destroying healthcare research
Sometimes, a tiny drop of malice could explain oceans of assumed incompetence

(Occam's razor beats Hanlon's razor)

atemerev··on Squeak 6.1
Wow, never heard of it! Thank you!
atemerev··on Chicken Scheme 6.0
Self-contained, compact, compiled, fast, ready to use.

SBCL is not compact and its "image" concept is not universally liked by everyone.

Guile and Racket are not fast (nor compact).

Chez Scheme is not "ready to use".

atemerev··on Squeak 6.1
And after all these years, still no fixes for high-DPI displays. The UI is still pixely and slow. Perhaps I should take Fable and try myself, this was the only bug I cared about and it is not solved after 10 years at least.
atemerev··on Qwen3.8 Max now ranked as the best overall model by agentic index
Well, that's the bad index then. It is barely usable in my opinion compared to other Chinese frontier models.
atemerev··on “Gravity is worth asking about”
Broken Windows Theory works.
atemerev··on Concurrency, interactivity, mutability, choose two
Choose three.

Preemptive actor model (Erlang, Elixir, Gleam) for backend.

https://pouchdb.com for frontend.

atemerev··on AI in Linux
Sure, and not only CO2 is a problem. Still beats burning coal and natural gas any given day.
atemerev··on AI in Linux
Well I see much less protests against gas and coal powered stations in the US than against, say, nuclear.
atemerev··on AI in Linux
Well thanks, but I prefer to have airplanes, and maybe occasionally even golf courses. As for climate change - you'd need to protest against Chinese (and Western, but mostly Chinese) coal and gas power plants way way before than you protest against airplanes and golf courses. I don't see too many protesters holding signs like "stop Chinese coal plants". For some reason.
← PreviousPage 2 of 34Next →