all things ai and coding while streaming, DM for consulting. pet projects @herorunai + https://xitter.cf/t.co/4lSXuoK33k
Joined January 2022
- Tweets7.2K
- Following729
- Followers2K
- Likes813
Pinned Tweet
I ran a 35-billion parameter AI agent on a $600 Mac mini.
Specs: M4 Mac-Mini 16GB RAM
The model doesn't fit in RAM. It pages from the SSD at 30 tokens/second.
On NVIDIA, the same paging gives you 1.6 tok/s. Apple Silicon gives you 30. That's 18.6x faster.
No cloud. No API keys. $0/month.
Here's what it can do 🧵
But @karpathy told me slack was the future of coding 😂😂😂
anthropic is massively underplaying the recent security breach. i'm hearing cathedral achieved full administrator access of the slack workspace (see every message, log in as arbitrary users) where literally every single research secret is contained
to those unfamiliar: Ant & OAI keep literally everything in Slack. it sounds exactly the same incident that happened to OpenAI in 2023, where OpenAI was hacked by a random guy who got access to an access token from an external app service
"we're critical to national security, we must beat China, and yes we keep all our crown jewels in Salesforce's Slack messenger. yes we have admin API tokens everywhere. why do you ask?"
thestreamingdev() retweeted
This is Genie 4, our most realistic and interactive world model yet. 🌐
Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak.
Why should all Sales Software cost money?
Introducing : Cooks is open-source revenue infrastructure for sales teams: sequencing, a parallel dialer, buying signals and call coaching. MIT licensed, on your own Postgres.
It is the category Nooks and Orum sell per seat, built clean-room from public docs as code you can read, run and change.
The parallel dialer rings up to four lines at once. The first live person lands in the rep’s ear with no beep, and machines are hung up. Cooks listens for the pickup itself and connects in about 0.8 s on our test carrier, against about three seconds for carrier answering-machine detection.
Sequences mix email, call, LinkedIn and manual steps. Sends go out in the prospect’s time zone, under per-mailbox daily caps, with spacing and jitter. A reply or a bounce stops the sequence for that person.
Signals put link clicks, replies, funding news, CRM stage changes and anything your bot pushes in through the API into one feed tied to the prospect and the account. Rules turn them into enrollments and tasks.
Coaching splits every recorded call by speaker, transcribes it and scores it on opener, discovery, listening, value, objections and next step. Talk ratio, longest monologue, one concrete tip. Reps rehearse against an AI prospect that interrupts and objects.
An MCP server lets Claude, Grok or Cursor look up the account, enroll the prospect, draft the email, place the call and log it. Suppression lists, unsubscribes, quiet hours and daily caps are enforced by the server on every send and dial, not by the prompt.
Hosted: sign up and you have a workspace. Self-hosted: Node 22 and Postgres, nothing else to run. Jobs, rate limits, webhooks and MCP sessions live in the same database as your prospects.
Github repo here, feel free to try it out and let me know what you think! Tag your favorite sales team
github.com/walter-grace/cook…
agi is going bring up a self religion - we will look back at all of the sacred texts because we are still going to always ask ourselves these questions as we increase or leisure from ai
Ahh yes but wework was stellar?
🚨🚨 @bhorowitz on Anthropic: "One of the most negligent companies that we've ever seen in terms of security."
Today, they filed to go public. Buyer beware.
thestreamingdev() retweeted
there will be within the next few years stunning advances in biology. there will whole new ways to think about what biological study even means. i would be only mildly surprised if there were near term solutions to problems like cancer, neurodegenerative disease, etc
LFG
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
I think the truth is come next year when most AI startups die, Jensen is going to be able to buy things at such massive discounts and is going to further monopolize Nvidia
JENSEN RESPONDS TO CONCERNS AROUND NEOCLOUD FINANCING:
"Everything else will get solved when demand is high. You can't solve the absence of demand. But when demand is high, everything else will get taken care of. I am quite confident that they will find solutions. Their solution starts with the fact that they have excellent customers. Demand is really high and they just need working capital, and working capital will come.”
$NBIS $CRWV $IREN $CIFR $APLD
everyone learning about state machines 🥱
@trello is the perfect orchestration pit for your agents. AGI is here and it is called TGI
thestreamingdev() retweeted
This illustrates the point I’ve been making about the limits of centralized approaches to SI safety in cyber.
OpenCode (and earlier Hugging Face) hit the same wall mid-attack: frontier models refused to analyze the very exploits needed for defense because the payloads tripped safety filters. The guardrails backfired.
Meanwhile, the attackers found a way around restrictions to access advanced capabilities. This is not surprising. Chinese models are only about 6–9 months behind the American frontier in cyber. Obviously the U.S. has no jurisdiction over them.
The only durable response is decentralized: make powerful models available so defenders can build and run their own cyber defenses. The only solution to SI-powered cyberattacks is SI-powered cyber defense.
