HNHacker News
TopNewBestAskShowJobs

airstrike

19,007 karma · joined December 2, 2010

hn at andyterra dot com
submissionscomments
airstrike··on Fox to buy Roku
my household and extended family has been running on roku for literally over a decade, in multiple countries, and not one person has complained. all of us, myself included, are perfectly happy with it
airstrike··on Anthropic's Safety Superpower
I think this just proves anyone can pick a benchmark that supports their point so maybe we shouldn't use treat them as evidence at all.
airstrike··on Swift at Apple: Migrating the TrueType hinting interpreter
Rust isn't named after what you think it is, but after fungi.
airstrike··on Noise infusion banned from statistical products published by Census Bureau
TIL Africa is a country
airstrike··on GLM 5.2 Is Out
> If they emulate reasoning well enough that it gets the same or better results what is the difference? Semantics?

No, of course not. The difference is that the ways in which we fail tend to be pretty ordered. You'd be hard pressed to find someone who's solved an Erdos problem but can't explain the difference between driving your car or walking to the carwash or can't count the Rs in their fruit names. Because if you can't count, you can't do math.

LLMs fail somewhat randomly because they do not have actual reasoning capabilities. It is hard to name that which they lack, because if we all knew, we would probably invent it.

Effectively, all problems are just search problems as" Newell and Simon argued as early as the 1950s. "LLM reasoning" today relies heavily on a side verifier. The coding loop that runs tests to see how it works, and so on. Which incidentally is what makes it so good at coding—that domain has a very quick and tight loop that can provide instant feedback about very targeted steps in their search.

But the corollary is LLM capability decays exactly along the gradient of verifier legibility. When you move to abstract problems that can't be easily verified, LLMs are pushovers with no real way to build nuanced abstract thought and literally think it through, find contradictions, decide on its own how to improve it and so on. They also have no spontaneous thinking, like you and I do in the shower sometimes. Because they have no agency, and those two things go hand in hand. Current transformer based models running on GPUs will never be efficient or fast enough to achieve that level of thinking. They're off by multiple orders of magnitude.

So the difference then is that their "approximate reasoning" is very useful, but is very flawed, and treating it as equivalent to human reasoning helps nobody. Believing in it is buying into hype, copium, and hopium. And, ironically, it likely delays the advent of proper AGI

airstrike··on Did Anthropic ask for this?
It's not about proving people are honest. It's about knowing that people either die a hero or live long enough to become the villain. Not because they want to, but because it's a natural journey given this competitive game of corporations.

OpenAI and Sam Altman being dishonest (or not) really has bearing on Anthropic. It's wholly unnecessary to argue about those other parties when discussing Anthropic or companies in general. Put simply, Alice being dishonest doesn't make Bob honest. But interestingly enough, OpenAI at first was also regarded as uncompromising stewards of "AI for everyone". It's in their name!

Speaking specifically about Anthropic, being concerned with AI existential risks is incredibly self-serving because guess who owns the very powerful AI that needs to be treated carefully and not allowed into the hands of other countries or competitors? All under the guise of being for the people. Even if a hundred people within the organization are doing this out of the goodness of their hearts, the company is not actually bound to that behavior and can renege on it at any point in time—just like OpenAI did.

Besides, Anthropic cofounders left OpenAI effectively forever ago. Just after GPT-2, which was pretty useless still. It was a whole different company. And the young are often naive but that naiveté eventually fades. Investors make demands, business realities sink in, and founders compromise.

This has happened time and time again. Or did you forget how Google Chrome started? Or how the shading and labeling of ads on Google has evolved over time? Or that "Do no evil" has been dropped?

The thing about my argument is that it doesn't require a purposeful nefarious motive. Corporations exist to maximize profits. When they act perfectly according top their stated goals, they make some decisions that are not in the interest of everyone else. It's important to remember that.

airstrike··on Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model
Worth reminding everyone that Lua was also created in Rio, though admittedly at PUC rather than by the government.

Rio has a strong engineering talent pool, along with many other major capitals in Brazil

airstrike··on GLM 5.2 Is Out
Yes, unironically claiming that and not wild at all if you're a practitioner.

It doesn't become actual reasoning just because you chose to call it so. If they did reason, LLMs would not fail at ridiculously easy problems like strawberry or car wash ones.

LLMs are great at search. They only emulate reasoning. They can't actually reason but they approximate it. Combine it with copious amount of computes and some search problems become tractable.

airstrike··on Did Anthropic ask for this?
We do not know for a fact what they genuinely believe, and many of us have seen companies act in opposition to their stated goals so the burden of proof is on them.

It's healthy to suspect ulterior motives from them.

airstrike··on Did Anthropic ask for this?
Anthropic had countless ways to fight this and they chose to cave.

The government can't apply export controls based on a control that does not exist. Creating one for model inference, if at all possible, would take 3-6 months at a minimum and it even includes a public comment process. That control is not cited anywhere because it does not exist.

The president can invoke emergency powers but that requires pointing to a specific foreign threat, notifying congress formally, posting on the national register, and it only lasts six months unless congress votes to renew it.

Given how easy it would have been for them to fight this, we can only conclude this was either outright designed or incredibly convenient for Anthropic.

Given their stated goal of pulling Fable by June 22nd, it seems likely they underestimated the amount of compute they would need or, even if they had perfectly estimated it, pivoting so that "the government shut it down because it's so powerful" on June 12nd is a better story than "we shut it down because we lack the compute" on the 22nd. This is especially true because the net new revenue from Fable is just from new signups between the two dates, which is likely smaller each day elapsed since the launch.

airstrike··on Not everyone is using AI for everything
ironically, I'd understand people not giving a straight answer on this particular topic
airstrike··on Rio de Janeiro's city government model Rio3.5 beats Qwen3.7 in recent benchmarks
how do you think that alleged amorphous mass of unaccountable bureaucrats got their jobs?
airstrike··on GLM 5.2 Is Out
I don't have a fully perfect definition, but I can name a couple of requirements.

Ironically, both reasoning and agency are required, neither of which our "reasoning agents" possess.

airstrike··on GLM 5.2 Is Out
Restricting access helps even less.

And none of this is AGI so...

airstrike··on Noise infusion banned from statistical products published by Census Bureau
it was a joke, to be clear
airstrike··on Open source AI must win
We have nothing anywhere near AGI/ASI so you're good for another 25 years, my friend
airstrike··on Noise infusion banned from statistical products published by Census Bureau
George Soros, probably
airstrike··on Noise infusion banned from statistical products published by Census Bureau
How do you get the idea that migration is "planned"? You've lost the plot entirely
airstrike··on Statement on US government directive to suspend access to Fable 5 and Mythos 5
Which, to be clear, does not mean they are actually honest.
airstrike··on Swift at Apple: Migrating the TrueType hinting interpreter
I'm not sure Rust has one specific use case as its main goal, despite being immediately suitable for systems programming.

I use it for making user-facing desktop applications, to name one example.

airstrike··on Swift at Apple: Migrating the TrueType hinting interpreter
These days I mainly write Rust but I did write a semi complex iOS app and enjoyed Swift. I just didn't love how slow the type checker was and how it got lost. I recall having to break things into smaller bits to help the compiler, and there were some oddities about the language.

The gap between the two languages is quite small, it just makes me wish Apple was also all-in on Rust

airstrike··on Slightly reducing the sloppiness of AI generated front end
You just need a few more smoke tests and you'll be fine.
airstrike··on Swift at Apple: Migrating the TrueType hinting interpreter
As much as I enjoyed Swift, one can only wonder what the world would look like if they had gone with Rust as their default language instead.
airstrike··on WASI 0.3
So join the community and be the change you want to see in the world?
airstrike··on CRISPR tech selectively shreds cancer cells, including "undruggable" cancers
I am not so certain
airstrike··on WASI 0.3
It's version 0.3...
airstrike··on WASI 0.3
Extending my Rust binary with a marketplace of WASM-based extensions like VSCode
airstrike··on Shall we play a game? My AI nuclear simulation
A grossly underrated movie. I think of it often these days.
airstrike··on Waymo Premier
Well, my wife was in an Uber ride where the driver asked her on a date, got rejected, and then started slaloming the car while complaining that he was so lonely he might as well drive the car into a wall ending his life, no wait, both their lives

She had to de-escalate the situation like a hostage negotiator would and assure him he was going to find love, and she was only not going out with him because she was already engaged, showing him her ring, but he seemed like a great guy

So, idk, maybe look the stats up.

airstrike··on Waymo Premier
This is the best combo possible. It feels like the ideal trade-off between the autonomy of car ownership and the flexibility of ride-sharing. I've been hearing about this concept for literally over a decade now.

I hope it expands quickly to the rest of the US.

← PreviousPage 10 of 34Next →