bigmadshoe

I took a six month break from AI coding after leaving my startup and just restarted this week. I feel like Opus 5.5 is a step function improvement from 4.5 or 5, which I built a large codebase with.

Opus 5.5 is capable of completing very long running tasks and managing many threads of work with subagents without getting confused in a way that simply wasn't possible before.

That isn't how we built things with lego when I was a kid. Try just skimming it and building something more random. It's super fun!

>That's got some nice But other than that, Mrs. Lincoln, how was the play? energy to it!

So you think that the move from 4o to Astra required something like 1000x the code to pull off? I find it hard to believe as an MLE, but since I don't work with LLMs I might be wrong about how complicated the implementations have become.

I do know people who work at OAI and it seems like none of them work directly on the models, but that's n=1.

I mean this sincerely: do you not use mo yourselves to test mo? It seems funny that basic bugs would get through.

This sentence is so badly written. What they were trying to say is that their product has much higher recall, slightly lower precision, and costs somewhere between Codex and Claude Code. You can see this by looking at the chart they included below, but I have no idea how this post made it to HN without someone noticing how confusing it is.

Are the buckets mutually exclusive?

Can anyone explain to me what this massive volume of code could actually be doing? They obviously have variations of model architectures, training infrastructure, serving infrastructure, and the source for their apps.

Why would the volume that any of this be increasing exponentially? And what are all the engineers doing when (besides the model quality) their products don't seem to be changing very much?

Reminds me of this: https://scontent-sjc6-1.xx.fbcdn.net/v/t39.30808-6/649087907...

That's fair, I've never worked in a domain where I need to run that locally. Is there a reason to not just SSH into a workstation that can run it?

Your point about the steam machine was correct at the time, but Linux gaming is flourishing now due to the steam deck. I think it's more accurate to say that the steam machine itself wasn't sufficiently compelling as a platform.

Why would you need a $6000 computer for dev work? For toying around with local models I would understand.

But why would we expect Anthropic to outperform the market over N years when they're opening at such a high valuation? It feels like most of the potential future growth short of world domination is already priced in.

Physically disconnect the computer the software runs on from the internet. People are overcomplicating this.

That's actually a really interesting idea. If we do end up in a "we don't look at the code ever" future, then something like this could actually be sensible.

Switch 2 actually still has lots of boxed games. Plus, when you buy games on the eShop you can use the virtual game card feature to gift them or lend them to friends after you finish playing. It's the closest experience to having a physical copy out of all the main digital stores.

I rewatched The Big Short recently, and all I could think about was AI labs.

This is already the world we live in. Lots of people are shipping code that they've barely looked at. Not saying I support it, but it's true.

Can you elaborate? Do you mean the model + harness combo reading more files than is necessary?

I would assume the context requirements are the same, i.e. whatever knowledge is needed to complete the task. There is no novel technique that can avoid this.

Any idea why you only got 160TPS when the blog post claims 1000+TPS? That seems like the only draw to this model right now while capabilities are behind the frontier.

What, in your opinion, is the point in solving these problems if no human is there to witness it? This is like suggesting we have AI watch movies for us and read its summary afterwards. In fact, why do anything anymore?

Yes, much of pure mathematics is "playing with puzzles" to the end of the joy and beauty revealed by a human solving them.

It should be investigated like other crimes. If someone was found to have intentionally created this situation, they would be held liable. In lieu of that, leadership should be held liable, or at a minimum the corporation should face massive financial penalties (on the order of XX% operating revenue).

This is super cool! You should put a clear example implementation of a feed forward network, transformer block, loss function, or something more ML specific on the homepage, assuming you're primarily targeting that domain.

EDIT: also the links to GitHub on https://lsalpha.com/systems are 404ing for me, so I exited your site without an example of what a real implementation might look like.

> Now, why did you just throw that out there without checking? Are you not interested in knowing the truth, you just want to argue for nuclear no matter what the facts say?

I'm genuinely interested, can you share the numbers you're referencing? It's surprising to me that a massive project like a nuclear reactor costs 1/3 of the inflation-adjusted total construction cost per year in upkeep. I'm not saying you're wrong, just the easy way to prove your point would be to provide some numbers.

I guess I need to read up on the law, but that seems like a wildly low bar for legality. If I hack into a system intentionally and unintentionally cause massive damages I'm free to go?

I think there is well established law around not running a program on your computer that hacks into systems. An LLM is a computer program. You're arguing that the ambiguity comes from the fact it was accidental? What about if I "accidentally" have an LLM convince people online to commit suicide, or drain bank accounts of seniors through hacking or social engineering. Where do you draw the line? It feels like your proposed reading of the law leaves "anything a computer program did without me intending it is legal", which is frankly insane.

We don't know the prompts but since these things often occur during "security tests" I think we can be fairly confident they did ask these systems to perform tasks that they knew could result in hacking. More transparency here would be appreciated.

Exports are typically ranked in dollars, not in kilograms. Do we think about the weight of GPUs a country exports or their value?

Why would the dropping value of the USD be a factor in selling gold? Doesn't holding gold hedge against exactly that?

I think the real reason is that they're afraid of political instability in the US that may result in 1) an inability to retrieve their gold, or 2) the government confiscating it.

2. is like saying that cars have specific laws around them, but sledgehammers don't, so murdering someone with a sledgehammer is inherently ambiguous. What? The computers being used belong to somebody who allows them to act this way.