HNHacker News
TopNewBestAskShowJobs

mattnewton

13,347 karma · joined June 16, 2013

matthewnewton.com
submissionscomments
mattnewton··on Claude Fable 5.2 Scores 65 on Artificial Analysis, Highest So Far
Right, I guess I mean why does the poster think more test time scaling means we’ve hit a “wall” versus just a different way to scale?
mattnewton··on Claude Fable 5.2 Scores 65 on Artificial Analysis, Highest So Far
Maybe a dumb question but once you have a model that spends a lot of tokens for high performance why can’t you distill it into a model that thinks less? Sure it wouldn’t generalize well to new tasks but plenty of tasks are important and recurrent enough to justify this.

Doesn’t seem like a “wall” to me.

mattnewton··on Every Windows 11 app should kill Electron, and Discord just showed why
That used to be true but with plugins vscode really has eaten the whole IDE workflow for most people
mattnewton··on Gemini Omni 1.2 Flash
Claude (and ChatGPT) has pretty good browser plugins that I use for integration with Google Docs. Between that when I am editing something on the tab and the mcp that lets it load files for context in chat I haven’t seen any blockers to my workflows, might be worth checking out.
mattnewton··on GPT-6 Astra covertly underperforms on safety research
It depends on how the safety is implemented. If it's implemented as a discrete step towards the end of training, or as separate classifier model or head on the base model, you're correct. But if part of the RL environment (as people are speculating here) or some earlier step, then there is no separate "unsafe" version.

A good example is in the Kimi k3.1 report, they identified that a lot of the cyber capabilities were emergent from parts of their pretraining codebase, and they removed those parts[0]. That means that they don't have a separate strong "cyber" Kimi k3.1 checkpoint that they then gimped, they trained it to be lower than they could have on this axis from the start and would have needed to go back to train on this and then redo parts or all of post raining to get the "unsafe" version.

[0] https://www.kimi.ai/blog/kimi-k3-1-pro#cybersecurity

mattnewton··on Kimi K3.1 Pro model weights are out
Zero Data Retention is the big one I think (and also the ToS don't allow multiple personal accounts for businesses anyways)
mattnewton··on Krea 3 Technical Report
Hi HN! This is Matt from the Krea AI team. We’re posting another large technical report of our most recent image model, Krea 3, which is focused on high quality edits at speed. Ask me anything!
mattnewton··on Evidence of a spying operation by ICE against left-wing social movements
> The fact you think all your political opponents are racists and evildoers is why the left lost in 2016, and in 2024.

I didn’t say that, I said that the current administration is an ally to racists and business owners who benefit from undocumented labor. This is just established fact at this point.

> Notice how it isn’t Australians or Germans ending up having to be rounded up by ICE.

Plenty of law abiding people who come here legally have been rounded up https://www.cbsnews.com/atlanta/news/south-korean-workers-de...

Which begs the question, why are white South African refuges welcomed by the admin while skilled work visa South Koreans are not?

But I think we agree more than we disagree- there should be a reasonable path towards citizenship that is rate limited. I think we might disagree on the rate limit and the reasonableness of the current path, but that’s an argument I would love to have over the current one we’re having over if the masked federal officers should be allowed to racially profile people and hold them at gunpoint without identifying themselves.

mattnewton··on Evidence of a spying operation by ICE against left-wing social movements
I agree, we should revise the laws to reflect what the people actually want- a reasonable path towards citizenship for immigrants working, paying taxes and not causing problems.

But that isn’t what the right wants; a modern coalition of racists who want no immigrants, and opportunists who want a underclass of labor that can’t organize, vote or otherwise fight back currently is in power.

And that coalition is in power after successfully scapegoating immigrants to average Americans who have no idea what the Byzantine legal citizenship process looks like now when they repeat that they want “legal immigrants”.

mattnewton··on Kimi K3.1 Pro and K3.1 Pro Highspeed
I’m not sure who this plan is for honestly, for general purpose LLMs in the Gemini-quality-tier how are you using enough tokens to justify $200 / month over just using deepseek over openrouter? And if you need a more capable model than that, like Gemini 4 argon, why not get a Claude subscription? Maybe the family sharing parts make it worth it?
mattnewton··on Investor-owned utilities charge 23% more for electricity than public providers
It’s not that controversial; usually I end up arguing with people who don’t believe utilities are natural monopolies rather than no government control. They usually point to starlink with the internet and solar + battery on power as incumbents that don’t need public easement. But that easement is too hard to compete with still imo and we ought to just have a public option.
mattnewton··on "I'm on a war footing": Terry Tao and the crisis in math
I don’t think you are going to vibe code your own bank or store website, you (or your agent) will still use their interface, no?

This is beside the point of the analogy though.

mattnewton··on "I'm on a war footing": Terry Tao and the crisis in math
An LLM automating web developers with 99% success rate saves the company millions of dollars and allows them to launch products and features faster.

A truck or a car that has a 99% success rate would be the top threat to society and quickly rise to the leading cause of death under 30 if widely deployed.

mattnewton··on Krea 3 – An image model built for editing at scale
The honest answer is that the team is pretty happy with the text to image of Krea 2 and our team prioritized editing which was the big gap in the previous model’s capabilities.

We know users are used to using one model for everything though and have already turned our attention to the text to image side; that message is to let people know that the text to image performance is not final, we expect the final tuning of Krea 3 to beat Krea 2’s high bar there soon.

mattnewton··on The gap in shared understanding of LLM capability is widening
I think this is related to the structures we’ve built in society. It’s lower friction to write software than just about any of the problems you listed.

It reminds me of the Ezra Klein quote that “OpenAl would need permits to cover its parking lot in solar panels, but it can accelerate into recursive self-improvement, as best I can tell, whenever it so chooses."

mattnewton··on Decisions API is in public beta
Why not try Jev?
mattnewton··on RuneScape's Position on Gen AI
There are lots of artists using generative ai in the same way there are artists using photoshop. It’s about control and intent, and it only gets weird with gen ai because you can give away both and still get results. Many people do take the role of a non creative manager, but that doesn’t meant you have to.
mattnewton··on GPT 6.1 Sol
Anthropic’s IPO?
mattnewton··on U.S. appeals court upholds designation of Anthropic as supply chain risk
I’m worried why many big companies aren’t worried about when the shoe is on the other foot.
mattnewton··on ZuckOff is a free app that sees Meta glasses before they see you
It has happened to me actually with the snap spectacle things.

Idk why we’re acting like people are hysterical the dangers of always on displays and cameras versus phones that go in pockets have been debated for over a decade now with Google glass. People do not like it for legitimate reasons.

mattnewton··on ZuckOff is a free app that sees Meta glasses before they see you
It is less noticeable in person than a phone out and pointed at you yes. And I would react similarly.
mattnewton··on ZuckOff is a free app that sees Meta glasses before they see you
Your phone has to be out of your pocket and raised up to record me, which I can notice.
mattnewton··on Qwen Image 2.1
It’s not hard architecturally, but it is hard to find or create good datasets of images on the magnitude you want. I suspect the qwen team heavily used synthetic data for this.
mattnewton··on Anthropic boss Dario Amodei calls for AI development to slow down
I don’t agree it’s that small. I think that the tools have advanced to the point where there are maybe 100 people pushing it forward at each company but a shortlist of thousands who would love to take their place and would be more than capable of it if entrusted with the compute and resources the current heads are.
mattnewton··on iPhone Duo
Android flagship vendors added their own wireless earbuds as well
mattnewton··on iPhone Duo
anyone with a pile of dongles disagrees here. It was clearly broken to sell AirPods
mattnewton··on Our decision on Cursor following its acquisition by SpaceX
Before, but they say that they’re also not going to provide astra to cursor subscribers
mattnewton··on U.S. State Department pauses immigrant visa applications
This is not that, these are people seeking to immigrate to the United States legally getting jerked around instead of a process that approves or denies them on a set timeline.
mattnewton··on AI Isn't Outthinking Mathematicians. It's Out-Remembering Them
Where would you get enough pre internet tokens to train a near-frontier model?
mattnewton··on US sells 30-year bonds at highest borrowing costs since 2001
I think deregulation of the banking sector at the very least contributed to the G in the GFC which congresses under Clinton and Bush did.
Page 1 of 34Next →