Hacker News: mattnewton

New comment by mattnewton in "U.S. State Department pauses immigrant visa applications"

mattnewton — Thu, 27 Aug 2026 14:57:14 +0000

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.

New comment by mattnewton in "AI Isn't Outthinking Mathematicians. It's Out-Remembering Them"

mattnewton — Sat, 15 Aug 2026 19:14:38 +0000

Where would you get enough pre internet tokens to train a near-frontier model?

New comment by mattnewton in "US sells 30-year bonds at highest borrowing costs since 2001"

mattnewton — Fri, 14 Aug 2026 14:33:33 +0000

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.

New comment by mattnewton in "US sells 30-year bonds at highest borrowing costs since 2001"

mattnewton — Thu, 13 Aug 2026 23:11:01 +0000

Congress sets the budgets (well, is supposed to, but yes there has been pretty excessive delegation to the executive) so you really want to say every Congress # + White House.

When you break it up that way, there have been several fiscally conservative congresses + good presidency combos, most notably under Clinton where they reformed welfare, increased taxes and managed to get a budget surplus one year. The formula seems to be slim Democratic Party majorities in Congress with a Democrat president.

So yes it’s rare but good governance + rising tides can make a difference.

New comment by mattnewton in "Mistral OCR 4.1"

mattnewton — Thu, 13 Aug 2026 23:05:11 +0000

I assume they pass the image alongside the text to redo/recheck the work. Yes this is silly, but is apparently required to get around refusals.

New comment by mattnewton in "Gloomberb"

mattnewton — Thu, 13 Aug 2026 15:54:02 +0000

TIL Bloomberg terminal is the original iMessage blue bubbles network

New comment by mattnewton in "Nvidia's Risky Business"

mattnewton — Tue, 11 Aug 2026 15:57:26 +0000

I think efficiency is unlikely to result in lower demand for compute, instead more useful compute per watt increases the value of that compute; and we are not going to run out of economically useful things to do with it anytime soon on the demand side.

The harder thing to forecast for me is if we hit a wall on increasing efficiency, either on the model weights side or silicon side, with current approaches. If we have to switch to something like burning the model weights into silicon to continue to make gains, then the current math on general purpose accelerators might be upside down.

New comment by mattnewton in "Qwen3.8-Max: A New Bar for Coding and Cowork"

mattnewton — Mon, 03 Aug 2026 16:29:15 +0000

If China keeps releasing LLMs with such permissive licenses it probably works better since the pretraining rnd is subsidized and de-risked - but that’s a big if.

New comment by mattnewton in "Qwen3.8-Max: A New Bar for Coding and Cowork"

mattnewton — Mon, 03 Aug 2026 16:26:04 +0000

The exception being the token embeddings and lm head (which scale with the number of tokens the model knows and presumably you need a smaller number in the tokenizer for only English and python). But those are a pretty small % of the total model weights on most LLM sizes

New comment by mattnewton in "Qwen3.8-Max: A New Bar for Coding and Cowork"

mattnewton — Mon, 03 Aug 2026 05:27:49 +0000

Honestly the 27b dense one punches way above its weight in a lot of domains, especially coding in my testing, so I think you will probably be disappointed.

New comment by mattnewton in "Qwen3.8-Max: A New Bar for Coding and Cowork"

mattnewton — Mon, 03 Aug 2026 05:26:26 +0000

I agree. 27b dense really did seem like the sweet spot.

New comment by mattnewton in "Qwen3.8-Max: A New Bar for Coding and Cowork"

mattnewton — Mon, 03 Aug 2026 03:50:38 +0000

There was a 3.5 122B 10A release -

https://huggingface.co/Qwen/Qwen3.5-122B-A10B

New comment by mattnewton in "Alphabet's cash burn raises alarm for Big Tech as AI spending climbs"

mattnewton — Thu, 23 Jul 2026 15:50:29 +0000

Because demand for inference tokens is above supply

New comment by mattnewton in "GPT-5.6"

mattnewton — Thu, 09 Jul 2026 20:23:29 +0000

This was a big concern for earlier models, but with modern CoT trained models they should be able to come to the conclusion entirely in the thinking trace.

New comment by mattnewton in "AI: The ROI Runway Could Be Long Outside the Tech Sector"

mattnewton — Tue, 07 Jul 2026 00:06:08 +0000

I think a) the labs are releasing very fast and b) why would they implement the long tail of app features when they can effectively sell tokens to every user to write their own version of the app, which is what is currently happening?

New comment by mattnewton in "Job seekers giving up: Labor force participation falls to lowest in 50 years"

mattnewton — Fri, 03 Jul 2026 20:11:36 +0000

It’s not that poor people can’t afford a stamp it’s that they aren’t going to spend money on an automated service or stamp if there are other places to apply to that don’t require this.

So I think it’s going to lower recall a a lot - it reduces volume of good and bad actors equally. Anecdotally, I’m not going to bother with such an application unless it’s my top choice or I’m desperate. For small companies that probably means throwing the babies out with the bathwater.

New comment by mattnewton in "Zuckerberg says Meta's bets on reorganization 'haven't come to fruition'"

mattnewton — Thu, 02 Jul 2026 23:13:49 +0000

Why do your think Meta doesn’t make money? Their ad platforms are incredibly lucrative.

New comment by mattnewton in "Job seekers giving up: Labor force participation falls to lowest in 50 years"

mattnewton — Thu, 02 Jul 2026 23:06:21 +0000

There are plenty of services to send mail form the internet for a small fee, so this will only discourage the most poor candidates and add friction for the best ones.

New comment by mattnewton in "Meta caps internal AI token spending"

mattnewton — Thu, 02 Jul 2026 03:04:27 +0000

I’m saying there is basically no way to both make vlms able to understand the long tail of PDFs where the layout conveys information (like charts and tables) and to make it as token efficient as text formats. Current approaches have mostly chosen to work more often than not at the cost of token efficiency.

New comment by mattnewton in "Meta caps internal AI token spending"

mattnewton — Thu, 02 Jul 2026 01:23:17 +0000

Because PDFs are a nightmare of a format and the only thing that’s is reasonably guaranteed about them is they will render to an image that people can read, the parsing of which will be much less token efficient than the equivalent text