SilverSlash

Yes, but open weights means you don't have to use their cloud providers. Someone in the US can host the model on their own infra.

I wonder if Anthropic just has better drafter models for speculative decoding and more efficient prompt caching in their agent harness.

Do you think the barbarians are at the gates of OpenAI and Anthropic? If cheaper, open weights models can seriously take revenue away from those two labs for (frontier - 1) model use cases (which are the models most enterprises will choose) then OpenAI and Anthropic are left only with users using their latest and greatest model AND who will keep upgrading to the newer ones?

Why will they have to raise prices eventually?

Fwiw I find social media scores corny

I agree. We've had large and thriving online forums for at least a decade before upvote systems became ubiquitous and things were just fine.

Is that impressive? If an LLM is just spewing out code which it has been trained on extensively then to me it's not that impressive at all. I want to see individual developers creating new software which previously took a whole team. Things not done previously. I am very pro AI btw, but the constant barrage of false hype is really tiresome.

One place where fiber cables cannot reach would be... way up in the air. Think about how many people fly each day and then remember how poor internet connectivity and speeds are at 40,000 ft.

So Starlink in flights seems like a perfect fit.

This was such a great read! Thank you! Too bad Oven Inc never got more headcount. Otherwise the engineers could've had a day hackathon week while the managers and founder went to a retreat for a strategy offsite.

Nah, it's a guy named Matt Deitke and he was 24 at the time.

So have they just given up on llama then? What happened to the 25 year old Zuck paid $250M for?

This is extremely misleading and not very useful. It makes little sense to use pricing per GB during decades when RAM was at most in MBs. In that case, why not talk about price per TB or PB? Then the line will look pretty much flat and horizontal.

I can definitely relate. My brain also wants to take the path of least effort now, even for simple things like adding two numbers which I could do very quickly in my head in my college days.

And with AI the path of least (initial) effort seems to be to just ask the model to solve it. It might get it wrong and then I'll prompt it again and again. But each individual prompt is fairly low effort on my part. Whereas coming up with the right solution myself might've taken less time but the initial effort is a lot more.

Last year I used to romanticize about building at least 1 thing each month completely by hand without any LLM coding help. The last such project I worked on was 6 months ago so sadly it's not going so well.

Difference is, humans learn from their mistakes.

Great! So next time the human will prompt the agent to watch out for and avoid this bug.

lol the new "swiss banks". store all your dirty data in digital swiss lockers

damn! then what's happens at the EU where the "goals of the organization" are themselves to increase and champion bureaucracy?

they already firmly in irrelevant territory

Very good. Just look at Gemma 4.

I like the concept. It would indeed be good to have a modern component library with AI design tropes as I think the old components libs haven't caught up. But in this particular case I must say, a lot of the components here just look plain ugly.

Useful concept but missing some obvious features like filtering on availability of tool calling or support for different modalities, etc.

I really like the agentic search feature that can keep you updated. Basically an ifttt but built into google search with an LLM on top.

Was this a Tesla with HW3 or HW4? Also, was it in the US or outside the US?

I hadn't heard of TanStack but a quick look at their website doesn't inspire confidence tbh. I mean, just take "TanStack Pacer".

It provides such things as:

```

import { Debouncer } from '@tanstack/pacer' // class

const debouncer = new Debouncer(fn, options)

debouncer.maybeExecute(args) // execute the debounced function

debouncer.cancel() // cancel the debounced function

debouncer.flush() // flush the debounced function

```

Why? Just why do you need to install some "framwork" for implement debouncing? Isn't this sort of absurdism the reason why the node ecosystem is so insecure and vulnerable in the first place? Just write a simple debouncer using vanilla js...

As in, htmx is better? I haven't used it but last I looked into it I was extremely confused as to whether it was a meme, an actual framework, or both.

A lot of the LLMs are very familiar with next.js and vercel is also aggressively building an ecosystem around their tooling for LLMs. So I wonder if this problem will only be exacerbated when everyone using LLMs is strongly nudged (forced) to use next?

Nope. I've been working on this project for a couple of days now and things were mostly going well. A significant portion of the mvp backend and frontend was built and working. Then this one seemingly simple bug appeared and just totally stumped both Codex and Claude Code.

There was even another UI component (in the same file) which was almost the same but slightly different and that one was correct. That's what I copy pasted and tweaked when I fixed the problem. But for some reason the models were utterly incapable of making that connection.

With Codex and Claude Code I thought maybe because these agentic coding tools are trained to be conservative with tokens and aggressively use grep that they weren't looking at the full file in one go.

But with Gemini I used the web version and literally pasted that entire file + screenshots detailing what was wrong (including the other component which was rendering correctly) and it still couldn't solve it. It was bewildering.

That's hard to believe in my case. I tried a variety of prompts, 3 different frontier models, provided manual screenshot(s), the agent itself also took its own screenshots from tests during the course of debugging. Nothing worked. I have now fixed the bug manually after 15-20 minutes of playing around with a codebase where I don't know the language and didn't write a single line of code until now.

I understand how laughable that sounds when I say it out loud. But the reality is, when I'm in a state of 'Tell LLM what to do, verify, repeat', it's really hard to sometimes break out of that loop and do manual fixes.

Maybe the brain has some advanced optimization where once you're in a loop, roughly staying inside that loop has a lower impedance than starting one. Maybe that's why the flow state feels so magical, it's when resistance is at its lowest. Maybe I need sleep.

Someone needs to make a compilation of all these classic OpenAI moments. Including hits like GPT-2 too dangerous, the 64x64 image model DALL-E too scary, "push the veil of ignorance back", AGI achieved internally, Q*/strawberry is able to solve math and is making OpenAI researchers panic, etc. etc.

I use Codex btw, and I really love it. But some of these companies have been so overhyping the capabilities of these models for years now that it's both funny to look back and tiresome to still keep hearing it.

Meanwhile I am at wits end after NONE OF Codex GPT-5.4 on Extra High, Claude Opus 4.6-1M on Max, Opus 4.6 on Max, and Gemini 3.1 Pro on High have been able to solve a very straightforward and basic UI bug I'm facing. To the point where, after wasting a day on this, I am now just going to go through the (single file) of code and just fix it myself.

Update: some 20 minutes later, I have fixed the bug. Despite not knowing this particular programming language or framework.

I'm deeply regretting paying for this service right now. There is some gaslighting going on in that issue that it's because of the 1M context model. I am using the non-1M context model and it's still disastrously bad.

Good, we agree :)