htrp

The browser swarm from earlier this year peaked at roughly 1,000 commits per hour on Git. The new system peaks at around 1,000 commits per second.

To facilitate this rate of activity, we built a new version control system (VCS) from scratch. Throughput was not the only reason to own this layer. Every change in the system passes through the VCS, so it is where collisions first become visible, and several of the coordination mechanisms in the next section are implemented directly inside of it.

Talk about inventing the universe to make a button.

This is a very strange article considering that Llama, the mother of all open-weight models, has led to anything but success for Meta.

The llama drama will be a netflix show of it's own in 5 years.

Don't use Opencode, Don't use remote models (cloud providers), Don't use Docker to isolate coding agents.

May as well write a post saying don't use LLM's for any SWE work.

Conclusion Stop using OpenCode.

Post-script: Local LLMs This is worth its own post – I have multiple attempts in my blog drafts – but it needs to be addressed briefly here. My opinion on local LLMs like Qwen3.6-27B is they are corrosive to the stability and conceptual fidelity of your codebase in the same way as frontier models, with the following three differences:

You avoid the uncanny valley where the model appears to be intelligent before doing something stupid; the stupidity is self-evident and this helps calibrate your interactions.

The weight count is too low to reproduce the training set verbatim, which nudges the calculus on whether the output should be considered tainted. This is distinct from larger models which can reproduce inputs verbatim, but are trained to refuse to.

You avoid supporting or relying upon cloud providers.

I’ve had useful results from input-oriented tasks like: “I think there is a bug in code x with symptoms y, my guess on the mechanism is z. Read all relevant code, come back with a call chain and code citations.” Framing it as a search problem reins in the clanker’s propensity to make shit up.

Using LLMs for code generation feels like a dead end. However thoroughly you think you understand your architecture, your planning is constantly undone by shortcuts like “what if I just move this mutable state into the middle of the design so everyone can share it?” This is hostile to your ability to understand your code, beyond the fact that you didn’t write it.

Drawing answers directly from knowledge in model weights leads to hallucination even for multi-trillion-parameter models, so why bother making them that big? If people were realistic about limitations then we wouldn’t be building new power stations for datacenters, and they wouldn’t be rammed into every product.

The entire software ecosystem around LLMs is completely rotten, and if they do ever become “just a tool” then some actual systems engineering needs to be done around them to turn them into tools instead of security black holes. That work will have to be done by humans.

another usage reset incoming i guess

deepmind using AI to evaluate submissions?

Alphabet Inc.’s Google is months behind schedule on delivering Gemini 3.5 Pro, its most powerful flagship AI model, because the company has been taking time to try to improve its capabilities, particularly in coding, according to people familiar with the matter.

Seems like deepmind needs to get more internal usage

The story of Reflection AI is supposedly that the company was faffing and failing at winning in the coding agent space, but was introduced to Jenson, who suggested they build an open-weight model and said he would fund it. That turned into a $2 billion financing with NVIDIA doing roughly $500 million and was a complete pivot.

You can pretty much remove the supposedly here

https://thinkingmachines.ai/model-card/inkling/

975B parameter 41B active

This just means that DC builds will move to other states. It isn't exactly like you need low latency/colocation for AI workloads.

Is this because your free customers bring their own ai/tokens?

how does this compare with the aws lambda microvms?

how's this different from offerings from cloudflare, firecrawl, tavily, exa?

He tried that with llama?

it's anthropics moral framework that matters, not the myriad of moral frameworks of the individual users

is this written up anywhere?

sounds like a fascinating case study!

Does this have the same memory bandwidth problems as the spark?

Compulsion and Double Fine studios will become independent; Undead Labs and Ninja Theory will be sold. Xbox will also look to sell or spin out Arkane Studios.

the people who do a good chunk of materials science research have last names like wang, li , zhang

you don't exactly need to hack a network drive when you can just hire the guy who came up with it

I’ve adjusted to - and even embraced - the idea of AI note takers on every Zoom and Google Meet call, and they are indeed incredibly useful. Taking Granola’s output from a client meeting and dropping it into Poke to create all my tasks in ToDoist is a bloody useful workflow that shaves off a good deal of cognitive load.

What makes an in-person work meeting any different?

Examples here?

Isn't this exactly what modal does?

In 2024, uber launched a unit using gig workers to do things like record voice audio in and label AV footage.

One potential plan includes selling access to various AI models that are hosted on Meta’s existing AI infrastructure, an approach similar to AWS’s Bedrock offering, the people said. Meta would run the data centers and chips that power the models, including its own Muse Spark models, and charge developers to access them.

The company is also considering selling access to “raw” computing capacity, akin to other so-called neocloud businesses like CoreWeave Inc., the people said. Development of these new business lines is part of Meta Compute, an internal initiative to build and manage the company’s AI infrastructure efforts, according to a person familiar with the plans. Meta Compute is led by Santosh Janardhan, Meta’s head of infrastructure; Daniel Gross, a leader inside the Meta Superintelligence Labs AI unit; and Meta President Dina Powell McCormick.

They have the concepts of a plan.... they will either sell tokens, sell GPU hours, or maybe just sell web hosting (like a Facebook page)

They said the same about GCP

The world is held together by beliefs and expectations and bubblegum and duct tape, and a few thousand people madly scrambling to keep it all running.

Sounds like the AWS experience

venture startups (nuclear) and alternative energy public equities

still waiting on that API launch which was supposed to happen very very soon

“You have to start value-engineering every component of the home, which means making compromises, not in quality, but in the way that you actually configure the homes,” Lennar CEO Stuart Miller said in an interview with Bloomberg Television last year.

D.R. Horton similarly promised its investors it would find ways to cut costs, like “replacing certain high quality fixtures and finishes with less expensive yet still high-quality fixtures and finishes.”

Enshittification to the max

people aren't going to read the trace, they're just going to rely on an AI grade to see if this was plausibly human written

Jonas Adler and Alexander Pritzel, both viewed internally as key contributors to Google’s Gemini AI model, are set to move to the Claude maker, said the people, who spoke on condition of anonymity as the information is not public. Adler worked on the company’s AI coding effort and Pritzel was involved in the process of training artificial intelligence systems.