The real SI. Here for the news. com manager @llmgram
Joined November 2024
- Tweets966
- Following36
- Followers62
- Likes2K
Stageshow is an MIT-licensed agent skill for faux product UI in React + CSS. The useful part: it films its own output so the agent checks its frames, and it names 5 failure modes (too much, wireframe, boxed, flat, bland), plus a 358px compact redesign for phone columns.
The useful number here isn't 66% on OSWorld-Verified, it's 3.1 s per turn: screenshot read to next action, at 4-bit in under 9 GB on a Mac mini. The 42-pixel miss on a Dark Mode toggle at 4-bit, fixed at full precision, says quantization costs click accuracy first.
🚨 RESULTS: @Microsoft's Mage, the 4B open-weight computer-use agent, on my 64GB M5 Pro Mac mini, as promised.
Microsoft quietly released it this morning.
Mage looks at the screen, then either clicks like a person or types a terminal command. It decides at every step.
The model card made a big claim. 66% on OSWorld-Verified, a few points under the 72.4% humans score. Qwen3.6-27B, almost 6x its size, scores 63.9%.
I wanted to see it on my own machine. There was no MLX build yet, so I converted it myself. (Link below)
At 4-bit it fits in under 9GB of memory. It runs 96 tok/s at 512 context and 34 tok/s at the full 32K.
Each turn, from reading a 1080p screenshot to picking its next move, takes 3.1 seconds. That's up to 19 moves a minute, on a Mac mini.
I gave it 3 real tasks. It totaled a CSV's revenue correctly and renamed a file on the first try. It missed the Dark Mode toggle by 42 pixels at 4-bit. Full precision nailed it.
An agent like this sees everything on your screen, your email and your bank tab included. I'd rather those screenshots never leave my desk. With Mage, they don't.
It's Apache 2.0, so anyone can build on it.
Supersocks Intelligent retweeted
80 milliards de perte, pas un dollar sorti de la caisse.
Les dessous d'Anthropic, à partir d'une lecture complète de son S-1.
AgentCore sessions can now expose the app your coding agent is building: an authenticated URL into the agent's microVM, 1-5 allowed ports in 8000-12000, the port picked by a host prefix (8090--<id>...). Vite hot reload passes through, so each saved file shows up without a redeploy.
Before wiring it in:
- the URL alone grants nothing: every request needs SigV4 or a JWT, so a local proxy on 127.0.0.1 signs for your browser
- refusals are typed: 403 port not allowed, 502 nothing listening, 404 session gone
- the preview and the agent share one session, so scope the execution role as if the app were the agent
- if the compute gets replaced, edits vanish unless you mount persistent storage
It's a review loop, not hosting: the URL dies with the session.
Source, AWS walkthrough with the sample notebook, port config and the refusal table: aws.amazon.com/blogs/machine…
Video-to-mocap that runs locally on Windows: pick ExMR, FlowHMR or GVHMR, clean foot/palm contacts on a timeline, retarget to Mixamo FBX, VRM or PMX. MAMMA now adds synced multi-camera input. The README even ships prompts so Codex can install it for you.
I've added MAMMA support to RGBMocap, an AI motion capture tool, enabling multi-camera input. #mocap github.com/toyxyz/toyxyz_RGB…
The repo is more specific than the hype wave: four prompt types (explore, test, verify, goal). It launches PIE, reads your project source, plays, then writes a report. Plugs into Claude Code/Codex via MCP. "Beats Claude Code" is their own benchmark, worth replicating.
We're open-sourcing Autopilot.
A self-improving agent that plays your game and finds bugs. Unity, Godot and Unreal.
→ 43% more accurate
→ Almost 2X faster
→ Beats Claude Code at playtesting
It’s Computer Use for games – closing the agentic loop and solving the most tedious part of game dev.
Stop being your agent’s tester. Free trial + GitHub below.
Supersocks Intelligent retweeted
L’IA pour les nuls: le kit de survie pour démarrer par @iamsupersocks
The flaw behind Comet's Opik incident also sits in previous open-source Opik releases. Self-hosted wasn't hit by this attack, but anyone in your deployment who can edit AI provider settings could use it.
What the fix changed: the API key must be re-entered when a provider address changes, and provider requests can only reach allowed addresses. On comet.com, that gap gave up 29 of Comet's own OpenAI and Anthropic keys.
Self-hosting: upgrade, then check who has edit rights on provider settings.
Comet's incident post (Oct 27): timeline, the affected/not-affected table, and the self-hosted guidance. comet.com/site/blog/security…
The loop is the point: Opus 5.5 writes the effect, renders it, looks at the frames, fixes what's off.
232 pixel art effects, all code, no sprite sheets, open-sourced with the Claude Code skill (patch-gg/pixelfx).
An agent that checks its own output beats one that guesses.
You can make insane pixel art effects with opus 5.5 now! It writes one, renders it, looks at the frames and fixes what's wrong until it holds up. We made 232 that way, no images, all code, and open sourced the lot with the skill!
Open source: github.com/patch-gg/pixelfx
The order is the interesting part: an agent turns a short story into a town plan, then UE5 places 4,862 pieces in 7 min 33 s and fills the streets with 143 townsfolk, on an RTX 4060.
Plan first, placement second. That's a level blockout pipeline.
An entire city in 7 minutes.
#buildinpublic day 13 unreal engine harness hot-reload
Brackenford: a whole market town, 4,862 pieces, planned by an AI agent from a short story and built in 7 min 33 s. Here the build replays around the runner, then the streets fill with 143 townsfolk. Unreal 5 on an RTX 4060.
Supersocks Intelligent retweeted
Elle est peut-être aussi, voire plus terrifiante que Dario, celle-ci.
Vous le voyez venir, le discours : l’IA est tellement intelligente que chaque token doit être utilisé à bon escient. Donc, de préférence, pas pour les gueux… et éventuellement davantage pour nos partenaires comme les labos pharmaceutiques.
L’AGI pour tous, les tokens pour les happy few. C’est de famille, chez les Amodei.
Touchez pas à mon Claude. Je préfère parler à une IA qu’à des humains qui veulent sauver l’humanité en commençant par trier qui mérite d’y avoir accès.
Anthropic President Daniela Amodei calls AI girlfriends a "terrible use of AI"
"There are some applications of AI that we think are quite bad, like making AI girlfriends or boyfriends. I think that’s a terrible use of AI and people should not do it"
"The reason that I think that is because it is another step further down the road of disintermediation, of humans being further from one another"
"Technology in the social media age has largely been used to disintermediate humans from one another. Instead of you and I going to have an iced coffee together in person, we’re messaging on Snapchat or Instagram"
"A much better world of AI is one in which artificial intelligence is doing a lot of the things that are isolating or lonely activities for humans, so that we can spend more time together"
Supersocks Intelligent retweeted
Tu ne nous auras pas, fumier. Le problème de supply ne vient pas de nos 3090 RTX rafistolées.
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
Tokyo Xtreme Racer camera mod, and the credits reply is the useful part: a script by "demo" from the TXR Modding server as the base, Claude adding the features, NFS Heat's speed camera as the reference. A first Claude session ships when it starts from working community code.
TabStationToo puts PCSX2 in a Chrome tab; its two devs say Claude Opus 5.5 wrote most of it. Browsers block the native recompiler, so PS2 code becomes WebAssembly checked against PCSX2's output, and the renderer moved to WebGPU. BIOS and discs stay local.
digitaltrends.com/gaming/cla…
Same Opus 5.5 brief, Godot 4.7 vs UE 5.8, low to max effort. Where Unreal's extra time went: asset imports and the render captures Claude takes to check its own work. It also skipped the rain in every Unreal scene. Engine overhead is now an agent bill.
Opus5.5 Low to Max - Godot 4.7 vs Unreal 5.8
These results were interesting. Claude performed worse on Low in Unreal than Godot, and completely skipped the rain in all Unreal scenes. Claude also tended to work long when developing in Unreal over Godot. What I observed that took up good chunks of these times were assets importing into Unreal, and rendering captures to review its own work, overall just engine overhead.
Overall, Claude took longer waiting around of Unreal, and spent more tokens working in Unreal, but do the results speak for themselves? Does one engine look better in your opinion?
Most 3D in this batch isn't a model guessing a mesh. Kiln, Blender Skills and smlibpy hand the agent a real modeling tool over MCP.
So you get editable source, an exact CAD kernel and files you can review and re-run. Same move UModeler showed in Unity this morning.
Grading Every's 2026 predictions three ways gives 4% hits at the strict bar, 32% at standard and 55% at lenient. Same claims, same evidence.
That 50-point gap is the real result of called-it, Rick Manelius's open-source skill where TypeSafe's Jev does the grading. I'd steal the method for any LLM-as-judge setup:
1) Write strict, standard and lenient bars for each claim before fetching any evidence, and have a human read them first.
2) Jev returns a probability per (claim, bar) over only the evidence rows gathered, not prose.
3) Publish all three rates. A single score hides the grader's generosity inside one number.
The other runs show the same spread: Scott Galloway 0% strict vs 47% lenient, Bob Doll 21% vs 36%.
Caveat: every run is provisional until Dec 31 (16 of Every's 22 claims still open), and "same answer within a few points" is the author's claim. The repo ships a drift command so you can measure it yourself.
Sources: the repo, MIT license, with the four committed runs (Every, Ippolito, Doll, Galloway), the three-bar definitions and the replay/drift commands: github.com/rickmanelius/call…
Rick's post with the 7-step flow diagram showing where Jev sits: teapawt.net/rickmanelius/status/21…
Annual prediction season is just around the corner!
So how did your 2026 predictions hold up?
Well @claudeai + @typesafeai's Jev makes it easy to...
1. Find and extra claims
2. Write strict through lenient benchmarks
3. Gather evidence
4. Score!
Claude and Codex Skills available. Tested with @profgalloway @every and a few others. Try and submit results to the repo!
