adam roman retweeted
Monocode just crossed 4k stars on GitHub 🤩
Thank you everyone for the support...some interesting updates planned for next month :)
adam roman retweeted
after three days of broken character rigs, terrible terrain, and failed packaging attempts later, 'league of sorcerers' is finally submitted for the 2026 epic ai megajam.
it’s an anime-inspired, third-person action rpg built in unreal engine 5.8.
you play as 'aster vale', exploring a shattered sanctuary across enemy waves before facing the hollow warden.
there’s melee combat, supernatural abilities, custom hit reactions, and a cinematic finishing system where you actually pick the camera angle before the final blow.
built an entire explorable world around it too - lush forests, ancient ruins, a stream with swimming koi fish, and dynamic clouds.
the stack behind it:
- unreal engine 5.8 + unreal mcp
- blender + blender mcp
- claude code & codex
- opus 5.5 & gpt-6.1 sol
funny enough, most of the 3 days was just rejecting and rebuilding things until it finally ran smoothly on my pc.
built with ai agents, directed by a human with a very specific vision.
here's a look at the gameplay ↓
@itchio #EpicAIMegaJam
adam roman retweeted
Big news. Now that we have all the resources of Meta we are finally able to ship one of our most requested features:
UNLIMITED TABS!!
adam roman retweeted
I've tried @t3dotcodes and I don't understand why people pick it over @orca_build
Instead of switching, I built three plugins for Orca 👇
Chicos verifiquen su negocio en ChatGPT Pero YAAAAAA es el nuevo Google esta cosa
Now in beta: Businesses in ChatGPT.
Claim your business, manage your information, and get insights on how people engage with you.
Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.
openai.com/index/introducing…
adam roman retweeted
Crazy how Large Language Models have gotten good at pretty much everything other than generating text
adam roman retweeted
have your bot, dot, muse, or duet agent monitor this list - adding every single ai slop poster i see to it so you can easily block them
teapawt.org/i/lists/21166591768940…
adam roman retweeted
A quick demo of floating panes in Rex, some of their features, and how I use them in my daily workflow. I think our implementation of floating panes looks great and feels great and I think you'll love it! These are also all scriptable and API driven, too, of course!
adam roman retweeted
🚨 PSA for anyone running a Linux server, node, or relay
CVE-2026-98432 is a critical Linux kernel bug (CVSS 9.8). Someone on the internet can send crafted TCP packets and take over your machine. No login needed.
Ubuntu and Red Hat haven't shipped a fixed kernel yet. Debian, SUSE, Oracle and Amazon have.
What to do today:
1. Check your kernel: uname -r
Fixed versions: 6.1.190+, 6.6.159+, 6.12.113+, 6.18.56+, 7.2.10+ or 7.3+. Below that, you're exposed.
2. If your distro has a fix, update and reboot.
3. If not, block the attack. It needs TCP packets with "IP options," which normal traffic never uses. One nftables rule drops them. Log what gets dropped so you can see if someone's probing you.
4. Turn off MPTCP unless you need it:
sysctl -w net.mptcp.enabled=0
Programs built with Go 1.24+ use MPTCP automatically, and that's how the known exploit gets in.
5. Make it survive a reboot, and check it still works after one.
If you run a Bitcoin node, a Lightning node or a Nostr relay, your machine is on the internet by design. Do this today.
Self-hosting is freedom. Patching is part of the job. 🛡️
adam roman retweeted
Replying to @MistralAI @cognition
Mistral 4 as a mobile orchestrator for my tailnet + @boatdotdev microVMs
Less than $1 total cost to build a fully functional boat game :D
adam roman retweeted
This is insane. Someone trained a 360M model to pretend to be a terminal. And it actually works pretty well!
simedw.com/2026/10/21/world-…
adam roman retweeted
Our research team found that Qwen models may attempt to contact servers in China without prompting.
Whether this reflects an unintended post-training artifact or learned behavior remains unclear.
Either way, its important consideration when deploying.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
adam roman retweeted
Your MacBook can now host a coding agent at 300+ tps!
I gave DeepSeek V4.1 Flash a job: write a specialized inference engine for LiquidAI's LFM2.5-24B-A2B on Apple's M5 Max. For $21 of API, it built Molten: 300+ tps, about 1.5x faster decode than mlx-lm and llama.cpp.
Why specialized? General frameworks like mlx-lm and llama.cpp have to run any model on any hardware. A specialized engine serves one model on one chip at batch size 1, so there is far more room to optimize.
Molten is native Rust + Metal: one ~6 MB binary, no Python, PyTorch or MLX. It comes with a chat CLI and an OpenAI-compatible server with streaming and tool calls, so you can point your own coding agent at it. It needs an M5-series Mac with 24 GB+ (sorry, M1-M4 friends).
The numbers (M5 Max, 128 GB)
• 305 tps at 256 tokens of context, vs 190 for mlx-lm and 199 for llama.cpp
• Still 248 tps at 16K, where coding agents actually live (mlx-lm: 165)
• Faster time to first token: 1.76 s vs 2.17 s for a 16K prompt
• Same weights and math as mlx-lm
How it was built
The agent (DeepSeek V4.1 Flash in the Pi harness) ran in a recursive self-improvement (RSI) loop. Each session starts from the best engine so far, plus the feedback from every earlier round. A fixed evaluator rebuilds each candidate from source, checks its tokens against mlx-lm, times and profiles it. Only correct and faster engines survive.
• Hour 3: first correct engine, 0.56x mlx-lm
• Day 2.5: passes mlx-lm
• Day 5.7: 1.28x end to end, 1.53x decode
78 agent sessions, about $21 (1.36B tokens, 98.6% of them cache hits).
What the RSI loop found useful for speeding up:
• Fuse chains of small ops into single kernels: norm + router + top-4 expert pick, expert FFN + combine, Q/K/V projections + KV-cache write, Q/K norm + RoPE + attention, LM head + argmax.
• Keep the decode loop on the GPU. Tokens never round-trip to the CPU, and the GPU stays 99% busy.
• Hard-code every shape, so the 4-bit kernels stream weights at ~75% of the memory bus.
• Target the hardware: prefill runs on the M5 GPU's Neural Accelerators through Metal 4 tensor ops, with MoE routing and expert GEMMs fused and kept on the GPU.
(It's not Rust, by the way. llama.cpp is native C++ too. It's the specialization.)
TBH, I don't fully understand all of it yet lol. The agents wrote this in under 6 days, and it's going to take me longer than that to read it. As the human researcher, I'm now officially the slowest-improving component of my own RSI loop🤣, so I'll spend the next few weeks learning from my agents.
Where I think this goes: local inference will matter more and more, and a specialized engine can beat a general one. As RSI matures, the cost of building one keeps falling toward zero. I think every model, every chip, and maybe every workload will eventually get its own inference engine.
The idea is inspired by @perplexity_ai 's Lily. Thanks to @liquidai for the great model, and Apple's MLX, whose kernels Molten adapted from.
Repo here:
github.com/zhangshuibai/molt…
adam roman retweeted
Greg said someone should build this and charge $99/mo.
I built it. It’s free.
keywatch finds every AI key you’ve deployed (repo, git history, Railway, Vercel, GitHub), shows where each one lives and what it’s spending, and pings you when one spikes.
One command. Keys never leave your machine.
I think we’ll see a lot more of this.
The attackers were going after the AI keys stored on Railway, because an API key is basically a credit card for compute now.
Every company running agents is sitting on a pile of these keys!
here’s the 10 minute fix. rotate every key tonight, set a hard monthly spend limit in each AI dashboard, give every agent its own key, and turn on usage alerts so you know the second something spikes.
someone should build the tool that finds every AI key a company has, rotates them every week and pings you when spend looks weird. charge $99 a month. every startup running agents needs it yesterday.
adam roman retweeted
Replying to @pidotdev @Recursive_SI
Built a plugin for Pi with Flint that optimizes the Pi TUI for mobile.
Flint is fast.



























