HNHacker News
TopNewBestAskShowJobs

nearbuy

1,512 karma · joined March 23, 2016

submissionscomments
nearbuy··on GPT-6 Astra
I think you're misunderstanding. Astra is at the top of the official ARC-AGI leaderboard, with an ARC-AGI approved harness. It's not a harness specialized for ARC-AGI. It just does the same thing the regular ChatGPT interface does: keeps conversation history across turns and compacts when it gets too long. Without the harness, it loses its entire context window every move. That's not how humans work and it's not how any real AI service works.
nearbuy··on GPT-6 Astra
People assume that's the reason because it's intuitive and "strawberry" is one token. But that doesn't explain why those models would also often get it wrong for "StRaWbErRy" or even "s-t-r-a-w-b-e-r-r-y", where the r's are not combined into one token.

We don't know what was going on inside the closed source GPT models, but this paper investigated on some of the open-weight models and found it's not due to tokenization: https://arxiv.org/abs/2604.00778

nearbuy··on GPT-6 Astra
I don't think we should count the lower tier models if we're discussing what the top ones are capable of. No one was suggesting that Sonnet is AGI.
nearbuy··on GPT-6 Astra
The last version to fail on those questions was GPT 4.5.

Meanwhile most humans fail to correctly answer how many f's are in the sentence, "Finished files are the result of years of scientific study combined with the experience of many years.".

nearbuy··on Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users
It also looks like they're saturating the test, with one LLM hitting the maximum possible score. (https://www.trackingai.org/home)

The test wasn't made to accurately measure IQs that high.

nearbuy··on Position: LLMs Can't Jump
The paper also fails to show that their central example, Einstein, relied on sensory experience for his intuition leaps rather than general reasoning. They just kind of claim that thought experiments require sensory experience. But you can easily ask an LLM to perform a thought experiment and simulate an outcome, and the SOTA LLMs generally seem to do about as well as a human. An LLM would certainly know that freefall feels the same as zero gravity, even if they haven't felt the sensation, which was the key intuition the paper talks about for Einstein's General Relativity. The paper's author would probably say any examples of this don't count, but without clear criteria for what would count, their claim is unfalsifiable.
nearbuy··on Qwen3.8-Max: A New Bar for Coding and Cowork
Only a tiny, tiny fraction of the parameters are encoding information that's specific to a particular programming language. Even if you could remove those without degrading performance, it would have a negligible effect on the model size.
nearbuy··on Qwen3.8-Max: A New Bar for Coding and Cowork
The problem is Claude Fable is now better than most programmers I know at software architecture and performance optimization as well.
nearbuy··on Gemini Robotics 2 brings whole body intelligence to robots
That's a good reason to use a dishwasher, and you should keep doing it. But the overall waste is small and people aren't going to care.

I'm not saying people who already have a dishwasher will throw it away. But a robot makes installing a new dishwasher much less tempting.

At my utility prices, it costs about 5¢ to hand-wash one serving (plate, pan, glass, utensils, and a bowl) with running water. That's not nothing, but it's dwarfed by the energy use of using the oven and the supply-chain energy and water use of the food if it includes meat.

People will casually waste 10x that energy without a second thought when they don't need to just because it's convenient, or because they're used to doing things that way.

The dishwasher looks more unfavorable if you're considering installing one in a small apartment or condo. Apart from the cost of the dishwasher, you need space to install it, and more cupboard space to keep more dishes, glasses, etc. For an extra 4¢ per person-meal, having your robot do the dishes buys you the convenience of skipping all that and having all your dishes clean and ready all the time.

nearbuy··on Is AI reasoning right for the wrong reasons?
Sure, but then there's no such thing as a network that isn't a classifier. Every physically computable function that terminates in finite time will map an input to a fixed set of outputs. And it goes against the common usage, where in machine learning we talk about classifiers, regressors, generative models, etc. as different things. They all become classifiers.

The parent commenter was trying to draw some insight from LLMs being classifiers that wouldn't apply equally to everything else.

nearbuy··on Is AI reasoning right for the wrong reasons?
The LLM has processed two data modalities derived from the apple (text and vision). Your brain processed a third (taste). But it is still just a data stream, sensing compounds and chemical properties of the apple and turning it into a stream of electric signals that reach your brain. You don't have any kind of ground-truth data stream that's inherently more powerful than what could potentially be fed to an AI.
nearbuy··on Is AI reasoning right for the wrong reasons?
LLMs are not classifiers. A classifier is an algorithm or neural net that assigns a label from a fixed set of labels to an input.

You can broaden the definition of classifier to anything that internally divides its input space into regions, but that definition would include every neural network, whether biological or artificial. So it's not very meaningful, and certainly doesn't give any insight into how they differ from humans.

nearbuy··on Gemini Robotics 2 brings whole body intelligence to robots
I have a dishwasher and I still usually just hand wash. It takes about 10 seconds to wash a dish.

The side benefit is all your dishes are always available. With the dishwasher, up to one full dishwasher load are dirty at any time, which means you need more dishes and more cupboard space than someone without a dishwasher.

If the robot is going to clear your dishes and wash them for you right away, what's the dishwasher for? To save the robot 10 minutes?

nearbuy··on Gemini Robotics 2 brings whole body intelligence to robots
Unitree's R1 humanoid robot is only about $6000, and it's still a nascent, smallish scale technology. They will come down.

If the future home robots are any good, it saves you from buying a dishwasher and robot vacuum. It can replace a maid/cleaning service and gardener/lawn-mowing service. For anyone paying for those services, it pays for itself.

$1000 isn't that high. Probably about a third of American households have an appliance over $2000, which is still a massive market.

nearbuy··on US citizen charged after GrapheneOS phone wipes during airport search
Yes, much like that. If he hadn't destroyed evidence, he could have argued it was malicious prosecution.
nearbuy··on US citizen charged after GrapheneOS phone wipes during airport search
What you're describing is malicious prosecution or abuse of process. It's illegal and it would destroy the prosecution's case. Not only that, but the victim could sue for damages.
nearbuy··on OpenAI’s accidental attack against Hugging Face is science fiction that happened
The irony is in this case the in-context and classifier "guardrails" would have almost certainly stopped the attack while their attempts at your definition of guardrails (the sandboxing) failed. In general, people keep trying to make secure systems and they fail with surprising regularity. Saying "they should have had better security" every time someone gets hacked is perhaps true, but it's not going to stop hacks from happening. And it's not a sufficient strategy on its own against future LLMs. "The Bitter Lesson" probably applies here.
nearbuy··on $100 AI Music Video: Claude Fable 5 vs. GPT-5.6 Sol
It's a strange experiment. Claude and GPT aren't generating the video. They're directing and editing it, and they request video from a generative video model using mainly text-to-video. Neither Claude nor GPT can actually watch video content, and the video model generates shoddy quality clips with artifacts, lack of consistency, and where actions aren't synced to the music.

It would be very hard for anyone to make a good music video with the tools Fable and Sol had available to them. They don't have precise control over the clips that get generated, and they can't see or hear the result, other than screenshots. So I'm not sure what the experimenters expected.

nearbuy··on It seems that the age of reading might be a short anomaly in human history
The UNESCO/World Bank literacy rate is basically defined how you thought. But high income countries don't usually report this because literacy by this measure is nearly universal. So they often report at higher thresholds (e.g. how many people can read at a grade 9 level), and news headlines often don't make it clear that this is not the same as the UNESCO definition.
nearbuy··on GPT-5.6 Sol Ultra will be in Codex
Questions like that cost a tiny fraction of a cent. "What's the capital of Sri Lanka?" cost a fifth of a cent at GPT 5.5 API price, and would cost a fraction of that if the question were routed to a more suitable, cheaper model. The output was 78 tokens.

By contrast, when coding, devs typically have hundreds of thousands of tokens in the context window, and may use many millions of input tokens per day.

Caching requires the full prefix to match exactly. If a single word differs near the beginning of the prompt, nothing after that can share the cache. So this type of caching would save a few queries that cost virtually nothing, but wouldn't help with the stuff where cost matters.

nearbuy··on GLM-5.2 – How to Run Locally
The usage is irrelevant if we're interested in cost per token. If you use it half as much, you get half as many tokens at half the cost. It's still $5.56 in electricity per million output tokens either way (using $0.20/kWh, adjust accordingly if you have cheaper electricity). If you use the API, you also pay half as much if you use half as much.
nearbuy··on Peopleless economy? Not technically impossible
Not much point in serfdom when they don't have use for human labor.
nearbuy··on How to earn a billion dollars
> I feel like you're also doing something weird—sorta strawmanning and sorta being conveniently inconsistent. You're reframing his argument as something much softer.

I'm going to firmly push back on this. A reasonable, straightforward interpretation of AOC's quote is that "you can't earn a billion dollars" without cheating or abusing others. I'm can believe she may have meant that as hyperbole, but even if she did, PG believes "What she meant was that it's impossible to get that rich without doing something bad — without cheating in some way." He's not doing mental gymnastics to get there; it's a straightforward reading of "you can break rules, you can abuse labor laws, you can pay people less than what they’re worth, but you can’t earn that."

That makes PG very much not strawmanning. He's arguing directly against her stated position. He doesn't steelman her argument (create a stronger or more generous version of her argument and argue against that), which could have made his argument stronger.

That also means I'm not strawmanning. PG's position is exactly what I said: that it's not impossible to become a billionaire without doing something bad. He is quite clear on that. Nowhere does PG use language implying all billionaires don't cheat. Only AOC does that, if you take her words literally.

I think you're basically doing what you accused me of doing here: greatly softening AOC's position. Did Dropbox spam users' contact lists to grow? (They didn't, AFAIK, though they did have a referral program to invite friends.) This is a huge downgrade in accusation of wrongdoing compared to breaking rules and abusing labor laws.

If you soften your definition of cheating to this level, it's no longer a billionaire thing. It's an everyone thing. Does my gardener report all his cash income? Has a friend ever streamed a TV show less than legitimately? AOC isn't trying to nitpick the smallest infractions. She's trying to paint becoming a billionaire as substantially corrupt.

nearbuy··on How to earn a billion dollars
You're not really engaging with PG's argument so much as nitpicking around the edges. Are people becoming billionaires primarily by cheating and stealing, or by making something people want? That's the core argument. Were GitLab, Dropbox and Stripe primarily cheating or offering a service people liked? Rather than cherry-picking one or two examples of what you deem the worst offenses, show that "you can't earn a billion dollars" without cheating or abusing others. PG isn't arguing that startups never do anything bad. He's just arguing that they can sometimes earn a billion dollars without cheating.
nearbuy··on Did Anthropic ask for this?
Anthropic's plan may not have great odds, but your proposal is orders of magnitude worse. There are plenty of groups opposing AI. They don't get billions in investment. They don't have a clear way to stop OpenAI or Qwen. They don't get a say in what values or safety measures the top AIs get.

You'd rather they signal their virtue and give up their ability to make a difference.

nearbuy··on Did Anthropic ask for this?
It means Mythos/Fable was the strongest model globally, not just the newest model from a company.
nearbuy··on Waymo Premier
That's not a reason for Waymo to give rides at a major loss. Heavy users are the ones that are going to buy the unlimited package. If you do just one average trip per day (there and back), at about $20 per ride, their regular price will cost you about $1200/month. A heavy user might use several times that. Do you think Waymo currently has 10-20x profit margins? If they're were able to drop their price by that much, they could also afford to drop their non-subscription price by 10x and take over the taxi/rideshare market.
nearbuy··on Waymo Premier
$100 per month is way, way too cheap for unlimited rides. That's about 10x cheaper than the amortized monthly cost for owning an average car, without the headache of doing maintenance. They'd be losing a ton of money on each subscriber.
nearbuy··on What it feels like to work with Mythos
These numbers don't mean anything without a denominator. You could burn $10 million/month of tokens if you want. We want to know how the cost per unit of useful output compares to a human. Does $6000 of usage buy you a man-month of work? Less? More?
nearbuy··on Nvidia is proposing a beast of a CPU system for Windows PCs
> Playing at 1080p doesn’t reduce your texture size, for the most part. You still use those 4k textures because you’re only seeing a subset of the texture projected at a close distance.

That greatly reduces your GPU memory bandwidth though. Sampling a subset of the texture only transfers that subset. Reading from higher mip levels uses less bandwidth. If your textures are high enough resolution to appear sharp at both resolutions (at least one texel per pixel), you need 4x more bandwidth to sample your material textures at 4k screen resolution for the same scene.

More importantly, material texture sampling is not most of your bandwidth to begin with. At 4k, most of your bandwidth is going to your full screen render passes. Especially with deferred rendering.

> DLSS isn’t just a magic on switch for free perfect up scaling. If you rendered at 720p and DLSS’ed up to 1080 it’s still going to look pretty rubbish.

I don't find this true at all. DLSS 4 Balanced looks excellent and renders at less than 720p for 1080p output.

← PreviousPage 2 of 25Next →