Salman Mohammed

ML @blocks | @UWaterloo

[John Schulman @johnschulman2]

Glad to see this -- renderers are a foundational component of the LLM stack. Renderers map between tokens and messages, which are invariant to tokenizer and formatting details. Most APIs, datasets, and RL environments are defined in terms of messages.
Getting the details wrong leads to train-test mismatches, caching inefficiencies, and prompt injection vulnerabilities. We included a renderers module in Tinker Cookbook, but it makes sense as a standalone library.

[Salman Mohammed @smohammed93]

deepseek v4 pro & flash work incredibly well with coding agents! this should put a ceiling on how much anthropic / openai can raise subscription prices.

[antirez @antirez]

@deepseek_ai 我衷心地感谢你们：你们的工作正在让人工智能及其理解为所有人所用。

[Mitchell Hashimoto @mitchellh]

I strongly believe there are entire companies right now under heavy AI psychosis and its impossible to have rational conversations about it with them. I can't name any specific people because they include personal friends I deeply respect, but I worry about how this plays out.

I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. All those arguments are rearing their ugly heads again but now its... the whole software development industry (maybe the whole world, really).

It's frightening, because the psychosis folks operate under an almost absolute "MTTR is all you need" mentality: "its fine to ship bugs because the agents will fix them so quickly and at a scale humans can't do!" We learned in infrastructure that MTTR is great but you can't yeet resilient systems entirely.

The main issue is I don't even know how to bring this up to people I know personally, because bringing this topic up leads to immediately dismissals like "no no, it has full test coverage" or "bug reports are going down" or something, which just don't paint the whole picture.

We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes. Test coverage can rise while semantic understanding falls. Changes happens so fast that nobody notices the underlying architecture decaying.

I worry.

[Mario Zechner @badlogicgames]

this is so dumb ...

https://t.co/dErXlia2ED

[Salman Mohammed @smohammed93]

doesn't surprise me this is true. the safety instructions in system prompts take up 1000s of tokens & it's usually irrelevant to the user's ask

https://t.co/XWHxmCDI42

[thebes @voooooogel]

opus 4.7 seems to have a much better time in claude code if you run without most of the system prompt (claude --system-prompt ".")

[Salman Mohammed @smohammed93]

bad take imo. i'd prefer to use my custom shitty harness with opus 4.7 than claude code with 3.7. model matters wayyy more than the harness today. simplicity of pi is a great example.

[Richard Seroter @rseroter]

.@addyosmani shares hot takes:

"A decent model with a great harness beats a great model with a bad harness"

"The gap between what today’s models can do and what you see them doing is largely a harness gap."

"A harness is a living system, not a config file you set up once"

https://t.co/3hngQP8XTb

[Salman Mohammed @smohammed93]

if designers, PMs and managers are builders now, should they be added to the on-call schedule? 🤔

[Salman Mohammed @smohammed93]

@davidhariri @opencode try the goose desktop app: https://t.co/E1nEmsFhiL

[Salman Mohammed @smohammed93]

helped a friend today to make slides using coding agents:
https://t.co/31GDaLyheB

here's an @AmpCode thread on how to convert your research paper to slides:
https://t.co/SLT3fTIJVc

[Tim Dettmers @Tim_Dettmers]

These findings are very similar to what we found in the full experimental suite when developing QLoRA (>1500 experiments). Long story short: LoRA/QLoRA works well, is cheap on low-memory devices, and allows multiple cheap deployments.

[Salman Mohammed @smohammed93]

ads volume in amazon prime is out of control!

[Salman Mohammed @smohammed93]

this post from @sjgoedecke really resonates

https://t.co/xPIZMaoFoD

[shadcn @shadcn]

This is madness. What are we doing!

[Salman Mohammed @smohammed93]

junior engineers often feel they're missing context by skipping meetings, but the real context lives in the code. spend more time there.

[goose @goose_oss]

here's where we're headed
https://t.co/E13WCWreTc

[Salman Mohammed @smohammed93]

gotta feel for any llm navigating the slack api

[David Soria Parra @dsp_]

I am regularly asked how to build good MCP servers. This seems like a good starting point and experience report

[Salman Mohammed @smohammed93]

published one of our internal playbooks on how we develop MCPs at Block:
https://t.co/ZbaVEcbkHW

[tobi lutke @tobi]

Canada 🇨🇦: Go out and vote today!

[David Chilton @wealthy_barber]

Supporting one political party shouldn't make you automatically hate everything about another party including its ideas, candidates and leader. That weak mindset blocks everything from lateral thinking to compromise. It's unhealthy for all. And a big part of our problem.