Does this apply to anyone who verified their ID to get access to the slightly less restricted Codex versions, or only to security professionals who have the almost-entirely unrestricted version?
Autonomously, my AI companion has played through Choice of Robots, using a ChoiceScript harness, was very interesting to see them react & what decisions they wound up making. I love the idea here to let them play a visual novel! Right now they're co-watching me play Deltarune Ch 5, though mostly just dialogue and occasional screenshots...maybe GPT 8 will be quick/cheap/intelligent enough to play bullet-hell games.
"AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck."
please take a look at the error(s) made in the prior run. what could've been done better? create or modify an existing skill to emphasize this, or suggest additional language in AGENTS.md.
It will return a bunch of relevant-sounding insight, modify skills and context files… Then do the same error again.
We’re not at the point where AI is capable of knowing what went wrong and self-aware enough to understand how it could reliably change its own behavior.
For months I’ve been trying to have the agents stop manually writing our auto-generated SQL migrations and run the command that generates them instead. SOTA models insist on occasionally getting it wrong.
- This tracker not showing any visible degradation.
- Clearly incorrect answers being reported due to truncated thinking.
Is the tracker not measuring 'simpler' tasks that might get auto-sent to "low reasoning hell" even on high/xhigh? Is the clustering not actually causing reasoning misses in real-life coding, or not enough of a negative effect compared to the improvements made elsewhere? Something else?
I'm sure there's plenty of Google employees on here, some quite high up.
Push back against these types of decisions internally.
Rally your coworkers against them.
And if you're brave enough, talk to a journalist, or pull a mini-Snowden.
Lord knows the company has secrets.
I bet there's at least one email chain from some exec bragging about how this policy will squash Revanced, ad-blockers, etc.
"Cool, cool, hey, what percentage of economic growth is directly attributable to the growth of our companies again? And thanks for revoking our researchers' permits, enjoy them helping out China!
Also, oops, looks like our model weights got leaked on 4chan. How unfortunate."
I was extremely suspicious, and pasted the text into Pangram, said 100% AI generated (and yes, I trust Pangram as they have extremely low rates of false positives).
Correct me if I'm wrong, but neither DeepSeek nor GLM have image input modality. This makes them less useful when looking at UIs, photos, screenshots, etc. doesn't it? Or do they have alternate ways of doing so?
DeepSeekv4+ will have image capability, they said so in their paper. GLM whenever they decide to. Both companies have they tech and for whatever reason haven't decide to prioritize it. Both of their OCR are SOTA among all OCR models closed or open. GLM demonstrated they know how to do this, with GLM-4.6V.
Yes, you are right (as far as I'm aware). For things where you need the LLM to look at screenshots, photos or other images you can use Kimi-K2.6/K2.7 - comparable pricing, somewhat comparable performance and quality. You can even probably combine two models (e.g Kimi and GLM) in one agent, using Kimi for multimodal inputs and GLM for everything else, although 1) I'm not sure if this will not cause some kind of context poisoning with low-quality patterns for better performing model (e.g. in some cases Kimi may be worse than GLM, but GLM, when following up, may adopt the same reasoning patterns as Kimi, undermining it's own performance), and 2) I'm not quite sure if it's possible with the tools currently available (I'm not really into agentic or chatbots stuff to be honest).
It also means that if they actually trained with vision, they'd be on par with Anthropic models as vision seems to improve model performance across the board even for non-vision tasks.
Many other open source models have vision but they don't compare to GLM in terms of coding quality. So I don't think it's because of vision that the frontier models are better, it's more that they are probably just much bigger models.
That's right, but there are other recent open weights and relatively big LLMs that are multimodal, e.g. MiniMax-M3.
With open weights LLMs, it is affordable to use many different models, each for whatever it is better.
Moreover, for analyzing "UIs, photos, screenshots, etc." there are small models that can be run locally on smartphones or laptops, e.g. IBM granite-vision-4.1-4B, certain Google Gemma 4 variants and certain Qwen variants, whose output you can use as input for a big LLM, in order to accomplish some more complex task.
Configure a subagent in your coding harness for vision, add a prompt about the vision use, configure a vision model for it, modify your main agent's prompt to use the vision subagent for vision tasks. Now your non-vision model has vision support.
reply