siliconc0w

Local interference really needs $50k before you're out of "toy" capability.

Given how subsidized the subscription plans are and the oss benchmaxxing vs real life performance- you're paying much more for far worse models.

Tinybox is cool but I think the market is maybe looking more for a turn-key explicit promise of some level of intelligence @ a certain Tok/s like "Kimi 2.5 at 50Tok/s".

Documenting my experiments using the 'Agent Flywheel' with my current project - https://clawfs.dev.

I reach for OpenCode + Kimi to save tokens on lower priority stuff and because it's quite fast on Fireworks AI.

A quick take documenting my experiments with the Agent Flywheel and the autonomous coding future.

All payer is really the most viable policy direction to save the US from medical bankruptcy.

I mostly do 2-3 agents yoloing with self "fresh eyes" review

It's not great that they found starlink terminals on Russian drones (they've since tried to lock them down more).

These should be export controlled and geo-locked as they are arguably much more powerful than any missile.

It's going higher, there is no viable TACO here and they can keep the strait shut indefinitely.

Plus the longer it goes, the harder it is to bring back up production or refining capacity.

Great insight and approach. I wonder though if instead of blogging this, he have the top labs bid on it - what that'd fetch?

https://taxmax.dev - AI tax agent that helps teams optimally classify their spend - we're looking for design partners if you want a free report.

The company's incentive is they get strong talent without the expensive of recruiters, they have a way to signal they're serious about hiring and aren't going to waste people's time.

A rejected candidate may burn the company but it requires burning their own stake as well so the incentive is really only there if they feel they've been significantly mistreated.

I can't even use 3.1 with Gemini CLI, not sure why.

Cyrpto had some interesting takes on these sort of problems that we haven't really applied more broadly.

The way I'd design a job board is require the applicant to escrow X and the job poster to escrow Y*X. Y is is the trust ratio. Given a bad experience, either the side can 'burn' the other and send both escrowed amounts to charity. An okay trust ratio might be 10, meaning they'll give you 10:1 burn ratio. A good one might be 100 or 1000. At that point they are essentially handing you a big stick to beat them with if they misbehave.

This would entirely eliminate spam and ghost jobs - suddenly everyone would be magically really responsive and polite.

Just a recent anecdote, I asked the newest Codex to create a UI element that would persist its value on change. I'm using Datastar and have the manual saved on-disk and linked from the AGENTS.md. It's a simple html element with an annotation, a new backend route, and updating a data model. And there are even examples of this elsewhere in the page/app.

I've asked it to do why harder things so I thought it'd easily one-shot this but for some reason it absolutely ate it on this task. I tried to re-prompt it several times but it kept digging a hole for itself, adding more and more in-line javascript and backend code (and not even cleaning up the old code).

It's hard to appreciate how unintuitive the failure modes are. It can do things probably only a handful of specialists can do but it can also critical fail on what is a straightforward junior programming task.

SCOTUS is entirely to blame for the chaos here, the courts quickly found the tariffs illegal but they used the shadow docket to stay the ruling causing the illegal behavior to continue for a year.

My experience is that the 'separate but equal' dual engineering track is largely a myth and that if you want advancement, the manager track is a much more viable track. Even with some of the recent flattening, there are still far more higher level roles for management than ICs. They are also given far more visibility and access inside the company which is extremely valuable in a large org. It also seems a good choice if you're not very good - I've seen bad managers hang around far longer than bad engineers.

They need to keep an emergency backup Claude to fix the production Claude when it goes down.

(More seriously I wonder if they'd consider using Openai or Gemini for this purpose)

+1 to Rust - if we're offloading the coding to the clankers, might as well front-load more complexity cost to offload operational cost. Sure, it isn't a particularly ergonomic or simple language but we're not the ones who have to use it.

If this wasn't ambitious enough, the author is also porting glibc to rust. As I understand it, all of it is agentic coded using custom harnesses.

Free chat users will get offloaded to the cheapest model heavily RLed to engagement hack and keep users on the platform. Maybe even paid plans if streaming is any indication.

I switched to Claude but the token efficiency and limits are much more noticeable. One or two coding questions and I'm at my session limit. And that is shared with chat too.

I was mostly able to get by with $20 codex but I'll probably have to splurge for the Max plan.

The problem with "Any Lawful Use" is that the DoD can essentially make that up. They can have an attorney draft a memo and put it in a drawer. The memo can say pretty much anything is legal - there is no judicial or external review outside the executive. If they are caught doing $illegal_thing, they then just need to point the memo. And we've seen this happen numerous times.

I actually prefer an OS-level API for Age verification rather than treating everyone as a child-by-default unless they upload their personal information to some random vendor.

BUT this is obviously not the right way to implement this.

Really interesting to see these new compute paradigms. I haven't built anything on Durable Objects yet but I can see the appeal and I'd prefer an OSS option.

SqliteDB per tenant may make sense, not sure about per actor. You really don't want to re-implement database transactions.

We need key AI researchers at these companies to speak out - execs will not care otherwise. If Jeff Dean made this a red line, it might matter.

I think the lower courts will easily find this is far beyond what the statue allows (and a free speech violation) but this Supreme Court will be eager to get back on Trump's good side and will stay the lower courts injuction.

That means at least a year to get it resolved, in the mean time practically all enterprise users will need to migrate off.

Google and Amazon both partner with them and sell to the US Government... so does this mean they can't run on Google or AWS infrastructure?

Good to them standing up to this administration. I doubt they actually want to put Claude in the kill-chain but this gives them a nice opportunity to go after 'woke AI' and maybe internal ammunition to go through the switching costs for xAI - given Elon more reason to line republican campaign coffers.

I'm guessing this is because Anthropic partners with Google Cloud which has the necessary controls for military workloads while xAI runs in hastily constructed datacenter mounted on trucks or whatever to skirt environmental laws.

you can tell he is a cracked out based tech ceo because he doesn't use capital letters