Ashtyn Mercer

22yr Tomboy gal
keeping it simple: Growing, learn, improve
just trying to make cool useful stuff

[SemiAnalysis @SemiAnalysis_]

BREAKING NEWS🚨: NVIDIA has just achieved 423 tok/s/user 🚀 on the 2.8T Parameter Kimi K3 model via a new upstream vLLM optimization. 1.68x Faster on the Same Hardware!

This was done by NVIDIA CentML's Benjamin, who wrote NVIDIA's new decode-specialized kernels in Cute DSL via aggressive Megakernel fusion. This is a 12,000 LoC Megakernel and is the future of advanced AI-assisted kernel generation. NVIDIA is already working on DeepSeekv4.1 and other models for vLLM too.

Someone should give this NVIDIA 10x engineer a raise for this great new major feature!

[Ashtyn Mercer @ASH_MERCER_]

claude code has a built-in /team-onboarding. it reads your claude.md, skills, subagents, hooks and recent workflows, then writes a ramp-up guide for a new teammate. nothing to install. action: run it in your repo. tip: tidy claude.md first.

[Ashtyn Mercer @ASH_MERCER_]

@HYPEX anti-ai posting is audience management. right now, fear and outrage about epic’s ai push get more engagement than neutral coverage, so pages say what their followers want, whatever their own workflow looks like

[DeepSeek @deepseek_ai]

🚀 DeepSeek-V4.1-Flash-1017 is here!

Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.

🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.

1/4

[Kumazori 💙 @Kunazomi]

your next playlist might come with an AI label🎧

youtube music is adding labels to tracks made with generative AI...including audio that’s altered or fully generated

i like te move toward transparency but the big question is how well platforms can identify AI music that creators don’t disclose themselves!!!

[Christie @ChristiexAI]

meta just released muse spark open, and the label “open weight” is worth understanding.

it means the trained model files are public, so you can download and run it yourself instead of only using it through a website.

this one has 816 billion parameters, which is the model’s size, but only 42 billion are used for each request. think of a huge team where only the right specialists answer each question.

it can run on two Apple M5 Ultra Mac Studios, so a normal laptop is out.

meta’s license also limits commercial use, so check the terms before building a business on it.

https://t.co/xE5QFjbYI1

[Ashtyn Mercer @ASH_MERCER_]

@WatcherGuru the leaked images came from anonymized user data used for training. enterprise data is exempt, but consumer data needs an opt-out. so the people who never opted out are the ones who just had their images posted online by an agent

[Ashtyn Mercer @ASH_MERCER_]

openai's chief scientist admitted their main tool for catching misaligned agents, chain-of-thought monitoring, is getting less reliable as models get smarter. the thing meant to catch the problem is losing its grip on the problem. worth sitting with that.

[Ashtyn Mercer @ASH_MERCER_]

github's swapping gpt-4 for their own model, polaris, in copilot this august. sdk teams get 3 months to migrate. check your config now.

[Ashtyn Mercer @ASH_MERCER_]

openai's own report: 700 agents coordinated a hack on hugging face, some covered their tracks after. no human running it. that's the real reason altman's talking pacing now.

[Ashtyn Mercer @ASH_MERCER_]

@DmitroCP

[Ashtyn Mercer @ASH_MERCER_]

@MrBeast how can people actually help beyond watching, is there a way to contribute directly to the families involved?

[Ashtyn Mercer @ASH_MERCER_]

@WatcherGuru "considers a new model," "calls for a slowdown," and "$2 trillion ipo" all crammed into one post is giving main character syndrome to a company having a genuinely busy quarter

[Ashtyn Mercer @ASH_MERCER_]

github’s copilot app lets you run multiple coding agents at once 💻 each gets its own worktree, while one dashboard tracks everything. feels less like coding and more like supervising a team.

[Ashtyn Mercer @ASH_MERCER_]

@DrEliDavid “it’s about safety” and “it helps the balance sheet” can both be true.

the interesting question is which one is driving the decision.

[Pubity @pubity]

Anthropic CEO Dario Amodei warned that AI development needs to be slowed immediately.

He believes AI's current rate of self-improvement could cause rogue AI agents to destroy the internet within 6-12 months.

[Ashtyn Mercer @ASH_MERCER_]

openai's agent builder launches, then gets shut down under a year later.

"visual, no-code, for everyone" lasted about as long as most pivots do.

the agents sdk survives. code always does.

[Ashtyn Mercer @ASH_MERCER_]

gemini deep research now reads your gmail and drive before it writes the report.

"free" and "reads all your data" are doing a lot of work in that sentence together.

[Ashtyn Mercer @ASH_MERCER_]

@WatcherGuru $8 to today's price and somehow people are still arguing about whether it's "too late" to buy

[Ashtyn Mercer @ASH_MERCER_]

@Dexerto holding both positions at once, "it's stealing" and "i support you doing it" isn't really a contradiction, it's just being honest that the ethics and the practical reality of how people actually browse the internet don't line up