building a game for my son for the last couple of weeks and it is so much fun
each day I add some features, watch him discover them, and update the agent for the next build.
seeing his reaction beats anything else I have shipped.
I add various math and writing challenges inside the game (to unlock chests, for example) and he loves that.
I use Godot and build for iPad.
For 3D models I use Qwen-Image -> Qwen-Image-Edit with the multiple-angles LoRA for four views -> Pixal3D
57
it was very effectively used in marriage counseling from late 70s till mid 80s
we dont know how anymore
This Post is from a suspended account. Learn more
Igor Tarasenko retweeted
Linus Torvalds says, rules prohibiting AI usage are “UTTERLY STUPID” and “idiotic”.
For the Linux kernel, Torvalds says, “we’re not going to make stupid rules that say ‘you can’t use tools for X’”.
the unit test defenders showed up in force.
not one stopped to think "maybe this person knows why unit tests exist and I'm the one missing something."
zero curiosity, maximum confidence.
we burn hours on this black-and-white garbage daily.
IMO the only tests you should have in the age of coding agents are, in order of priority:
1. Full E2E tests. Nothing mocked out at all. Can even be something that runs in prod on test accounts via Playwright or equivalent.
2. Integration tests. Agents make more mistakes as things bleed between data/API boundaries where schemas can drift.
3. Golden tests. This helps ground the code with real examples of data, and can be used as regressions against edge cases.
Unit tests are bloat 99% of the time if you’re using a frontier model since they’re smart enough now to get the implementation (and subsequent iterations) right on the first shot, so they only add bloat at this point.
1
124
Cloudflare’s Jev-like model matches Jev’s performance at twice the speed. At least. It understands images and videos, too.
1
60
in a week we'll be doing software engineering in Jev.
extract primitives from the codebase → build with Jev → loop
1
4
105
Nice how he turns rumor "Anthropic resolved a major open problem" into a story about OpenAI with "drop the Anthropic author" "ruin your career" "don't have to be nice"
Anthropic gets proof credit without owning the rushed papers or fight
"I am not accusing anyone of anything"
I deleted my previous post. I'm just going to post the entire statement page by page. It's a complicated situation.
Statement: cims.nyu.edu/~tristanb/state…
cims.nyu.edu/~tristanb/euler…
cims.nyu.edu/~tristanb/ipm.p…
cims.nyu.edu/~tristanb/bouss…
github.com/tristanbuckmaster…
2
135
for many months there was nothing close to gpt pro model, and still there is nothing!
especially for planning development of complex software.
i am surprised people only now start to use it.
1
114
h3 from @MiniMax_AI is a big deal.
the gap between open and closed video gen models is way bigger than what you see with LLMs. h3 makes that gap a lot smaller.
And video gen is expensive.
So being able to run SOTA video gen that's completely uncensored, actually knows about copyrighted material, and lets you train LORAs - is huge!
Just on price alone you are looking at at least 10x savings over the video gen APIs.
1
66
pangram always used loud claims and misleading ui to convince people that such detector can be reliable
but i think that it can get to the point where errors on "AI generated" claims will be close to zero
of course errors on "human written" will never be close to zero and will actually only grow
i don't think that any serious decision should be ever made based solely on the results of this tool
1
187
Opus's issue for me is the increasing interface cost, which means I spend more time managing the model instead of focusing on the task.
It has more personality, leading to more second-guessing about what actually happened and less tool-like execution.
1
3
793
I don't buy this "distillation" through API giving significant results.
Yeah, it can make a model talk in the parent model's style, maybe even perform better on certain tasks.
But the importance Anthropic is pushing on everyone is just ridiculous.
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection. Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.
The United States strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.
2
105
Igor Tarasenko retweeted
The ADHD dream career is being a highly paid consultant who swoops in, solves a complex crisis with a chaotic stroke of genius, and immediately exits before having to fill out the timesheets.
Real breakthroughs happens when you move beyond seeing AI as just a magic answer box and start engaging with it as a system that still relies on your judgment.
2
83
just in time. open source models lately have been surprisingly underwhelming, and there's been a lot of debate about whether they're actually catching up to closed models or falling further behind.
i'm always curious how much safety and censorship alignment is quietly eating into raw model performance.
We're introducing GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a solid 1M-token context. GLM-5.2's new capabilities include:
Solid 1M Context: A solid 1M-token context that stably sustains long-horizon work
Advanced Coding with Flexible Effort: Stronger coding capabilities with multiple thinking effort levels to balance performance and latency
Improved Architecture: We propose IndexShare, which reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at a 1M context length. We also improve GLM-5.2’s MTP layer for speculative decoding, increasing the acceptance length by up to 20%
Pure Open: An MIT open-source license — no regional limits, technical access without borders
Supporting long-horizon tasks starts with making long context engineering-usable: the model must maintain quality across long, messy coding-agent trajectories, not just accept more tokens. A 1M context is easy to claim, but much harder to keep reliable under real engineering pressure. To this end, we substantially expanded 1M-context training for coding-agent scenarios, covering large-scale implementation, automated research, performance optimization, and complex debugging. The result is a long-context system that is not only wide in scope, but solid in execution: a practical substrate for sustained engineering work.
This capability is reflected in GLM-5.2's performance on three long-horizon coding benchmarks. FrontierSWE measures whether an agent can complete open-ended technical projects at the scale of hours to tens of hours, spanning systems optimization, large-scale code construction, and applied ML research. On this benchmark, GLM-5.2 trails Opus 4.8 by only 1%, while edging out GPT-5.5 by 1% and Opus 4.7 by 11%. On PostTrainBench, where each agent is given an H100 GPU and evaluated by how much it can improve small models through post-training, GLM-5.2 outperforms both Opus 4.7 and GPT-5.5, ranking second only to Opus 4.8. On SWE-Marathon, an ultra-long-horizon software engineering benchmark covering tasks such as building compilers, optimizing kernels, and developing production-grade services, GLM-5.2 still has room to grow, trailing Opus 4.8 by 13% while remaining second only to the Opus series. Across all three benchmarks, GLM-5.2 is the highest-ranked open-source model, showing that its 1M context has translated into practical long-horizon delivery capability.
1
3
123


















