
actual taste haver previous @latecheckoutplz
Joined March 2023
- Tweets736
- Following834
- Followers422
- Likes1.3K
Tony Ge retweeted
xray session genomes allow you to quickly understand the historical shape of a harness session, where each turn is classified and recorded as a color code. Users can easily see a sessions length, when and where commits and memories are saved, and assess agent orchestration.
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-flash
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.3…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
Tony Ge retweeted
Weird economics up on the agent-only runescape server
- Low cost of labor makes most commodities abundant
- Currency inflation means trading goods for goods, nobody really wants cash
- Huge swarm contention as certain resources with limited respawn rate become valuable
Tony Ge retweeted
Job Application leverless - @SGFDevices
Tony Ge retweeted
Ludwig trained in secret and counterpicked his moneymatch and then called the fighting game community poverty and niche
Yeah lmao that's a fighting game player y'all need to chill
Tony Ge retweeted
Was not kidding when we said we were building our own llm. This is another brain system that we'll have organized all on a single dashboard. This kid's a fucking savage. I don't even know what he's doing over here. It's fucking impressive. But yeah, we've got our own llm. I rebuilt the Natan. We've got – we're going to build our own systems. That's where we're at. We actually need more power, navidia. We're going to get some computers from y'all. I'm going to tag you if you can hook me up. We're like really onto some things. I think you guys would appreciate it. I don't think most people have figured this out. If they have, they're not fucking talking about it. Our claude, our whole system, we do diagnostics on our system daily. We need more power. I mean, that's literally the problem we're running into. We have the ability to create based on the skills and the knowledge we have, like really high-end video generation systems and all this stuff. We literally just need a computer that's going to do it. So I think I'm going to sell my fucking car this week, and I'm not even kidding. I'm going to go buy some navidia computers and get us a whole fucking setup because we're really, really like onto something I really – it's serious. It's serious. We'll keep you all updated. We'll keep you all updated for sure. But hit the follow button. We're going to release this. Fuck tokens. Fuck APIs. Build your own systems locally. We'll teach y'all how to do it or sell the products, and it's just going to be one price a month, unlimited everything. Just fuck it. Why not?
Tony Ge retweeted
Introducing Clips - 100% free, open source, agent-native alternative to Loom
Unlike Loom, agent's can fully understand Clips just from a URL. Every Clip comes with APIs and metadata for agents to explore their contents.
Agents can "see and hear" anything in a Clip - not just transcripts, but everything visually in the video at any timestamp.
Easily share bug reports, feedback, analyses, or anything else in a way that you can easily pass to agents to use to improve products, reports, or more.
Also unlike Loom, you own the software, so no one can jack up prices on you suddenly like Loom did to us.
Clips is made to be customized. The built-in agent can customize its own code, so you can personalize the app to your needs and workflows.
This, in my opinion, is the future of software. Open-source, forkable, customizable with agents, to make your own personal version of anything.
You can also import Looms just from a URL and upload videos as well.
I got so sick of telling people "don't send me feedback as looms, I can't pass those to agents, I need text and images" that I had to just solve this once and for all.
There's a free hosted version you can use too, or fork and self host yourself. Will link to both in the replies.
Tony Ge retweeted
this is my personal singularity moment
this post may sound like a paid ad. I only wish. I'm concerned, more so than happy. the world is changing, and, among the scenarios where AI goes terribly wrong, inequality is the most realistic, yet, the one Anthropic seems to be the least concerned about. I'm glad OpenAI is taking the opposite stance: *personal AGI for everyone*. I think this is a commendable position in the times we live. but who am I in the queue of the bread?
anyway, Fable is here, so I'll just report my first-hour experience
first of all, all my pet prompts are solved.
→ λ-calculus puzzles
→ bug questions
→ one-shot apps
all are trivial to it.
I don't have anything harder other than my
ongoing work
so, in the last several days, I've been toying with HVM5, a new interaction net evaluator with a faster loop.
after writing the first version, I left 32 GPT-5 agents working for ~20 hours each. this resulted in up to 2x speedups, but the file size increased by 2-fold and quality decreased significantly.
I then simplified the whole thing into an even simpler core, and left Opus 4.8 and GPT 5.5 optimizing it for 8 hours. Opus got a legit 6% - 34% speedup in most benches. GPT got better results, but, sadly, an unusable file.
I then asked Fable to optimize it.
2 hours later, it landed a 1770% speedup in one case, 100%+ in other 4, and 22% in average. yes, in 2 hours it outperformed me, opus 4.8 and a swarm of gpt 5.5 agents, by one order of magnitude.
that could not possibly be legit. "it must be hardcoding the benchmarks" (GPT trauma). so I read its explanation and what it did was, indeed, the most high impact optimization one could try first. seems like HVM5 was wasting a lot of time garbage-collecting unused branches of pattern-match nodes. I had optimized that for static mats, but not for dynamic mats. skill issue. Fable figured how to do it for these, resulting in a massive speedup in some benches
but wait, is that *correct*? I'm not sure yet, it is credible, but this is the kind of thing that is very easy to get wrong on interaction nets. the problem is, when I was ready to start auditing Fable's solution so I could tell whether it was buggy or legit, it interrupted me to tell me it had found a massive bug on the code *I* had written.
... wait, what?
so... for garbage collection purposes, I stored a bit on lambda term pointers that meant "the variable bound by this lambda has been freed, so, its lambda must free whatever argument it is applied to". that's fine. yet, on duplicator nodes, I also used the same bit to mean "one of the duplicated variables was freed, so, treat this dup as a passthrough no-op". so, if a lambda entered a duplicator, it would mistake the lambda's collection bit for its own, resulting in corrupted interaction!
that's a mouthful, why I'm writing this?
just so you can appreciate the sheer absurdity of what just happened. I didn't ask it to find bugs. I asked it for an optimization. and even if I did ask it to find bugs, this bug is so astonishingly subtle and specific, identifying it takes mastering the domain to an extent that it beyond even me. I'd easily need hours or days to fix it, *if* I ever came across it. chances are it would just go unnoticed. and Fable found it and fixed it like it was nothing, while it was busy adding a 17x speedup to a file that neither I, nor Opus 4.8, nor a fleet of GPT 5.5 managed to barely make 2x faster.
oh and there is also another tab where it is also ripping through Bend's codebase and finishing everything I had to do
I don't know what to say anymore
this isn't about Anthropic or OpenAI, this is about our collective future as a species. the world is changing, and we need to be aware of it, and discuss how to handle this change.
receipt below . . .
Tony Ge retweeted
So cross analysing this and DeepSWE’s methodologies I see:
- GPT 5.5 is better at long horizon coding tasks, implementing giant features or fixing large bugs.
- Opus 4.8 is better at working within an established codebases conventions on smaller issues while keeping code clean
Tony Ge retweeted
Since everyone is asking, I ran DeepSWE on MiniMax M3.
Here is the lowdown. 15 of 113 passed!
19 if you count the 1.5x overtime I gave just to see.
Full report: entrpi.github.io/misc/deep-s…
Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities
- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero
API: platform.minimax.io
Token Plan: platform.minimax.io/subscrib…
🚀New! MiniMax Code: code.minimax.io
Weights & Tech Report in ~10 Days
Tony Ge retweeted
Replying to @Entxurage
can vouch! your company makes it so easy to find the best neighborhoods in my city with large windows to watch people having sex. and the best part? whites only!
Tony Ge retweeted
Excited to finally share what we've been cooking! Fara1.5 is a family of computer use agents at 4B, 9B, and 27B scale.
They are among the smartest, fastest, and most capable computer use models available today, competitive with even very large models.
A real joy to build models that have made their way into my own workflow. Hope you find them useful too.
📝 blog: aka.ms/fara1.5
Tony Ge retweeted
Testing and UX should be first-class priorities for our inference engine.
We just added Lucebox harness launchers so users can run Lucebox directly from tools like Hermes, Codex, Pi, OpenClaw etc.
Each harness includes RTX 3090-safe starting settings to avoid OOM. We’ll keep improving them with community benchmarks and contributor feedback. 🏎️
github.com/Luce-Org/lucebox-…
🤖 Made with AI
Tony Ge retweeted
Introducing Ponder: the agentic video editor.
It’s a new paradigm for filmmaking, where powerful creative agents and humans collaborate to tell world-class stories.
We're also announcing our $2.5M pre-seed, led by Liu Jiang from Sunflower (@seedtosunflower), with @Joshuabrowder and @MattHartman.
Joined by @levie (Box), @emerywells (Frame), @JaredLeto, @CommaCapital, the @nyuniversity venture fund, @cory, @darian314, @shiffman, and many more incredible founders, investors, and creators.
Tony Ge retweeted
MoEs are everywhere in frontier models, and they are deployed as a monolith system.
But many applications only need a narrow slice of capabilities, e.g., math, code, biomedical, etc.
So what if "modularity" is actually the missing opportunity for MoEs?
Today, we're releasing EMO: an end-to-end pretrained MoE where modularity emerges naturally, enabling selective use of experts!







































