I have the theory that agents will improve a lot when trained on more recent training data. Like I‘ve had agents have context anxiety because they still think an average LLM context window is around 32k tokens. Also building agents with agents, letting them do prompt engineering etc, still is very unsatisfactory, they keep talking about GPT-3.5 or Gemini 1.5 and try to optimize the prompts for those old models, which of course was almost a totally different thing. So I‘m thinking if that‘s how they are thinking of themselves as well, maybe that artificially limits their agentic behavior too, because they just don’t know how much more capable they are than GPT-3.5
That‘s entirely true, but my entire point is you cannot judge somebody as incompetent based on something they are not trying to achieve. You can say he‘s a bad president, makes sense because a good president serves his country. But you can’t say he is an incompetent one just because he has ulterior motives.
I‘m not making a moral judgement here. I‘m saying even if the attack happens for different reasons than those you have been told (honestly a US classic), the search for a coherent reason behind it (no matter if justifiable) cannot end by saying „It doesn’t serve us. He is a stupid president, he promised us something and did something else he’s so incompetent“ because incompetent has to mean in regards to something. And if his goal really is narcissistic validation and money, well then he surely either is very lucky currently or not incompetent.
No that’s not what I mean. I‘m saying we should acknowledge them, not necessarily by doing so charitably. My point is that you cannot have productive debates about politicians if you think they all act to benefit their people and if they do something that contradicts that suddenly everyone says they are incompetent or stupid. Only then we can start asking what is he trying to accomplish with that. Which could be an interesting debate. What is he trying to accomplish for US citizens with that is a stupid question
I don‘t disagree with that, in fact that’s exactly what I‘m saying. If someone becomes president to achieve those goals, do you really call him incompetent for building his cabinet around achieving that very goal?
Were they really campaign goals or were they just marketing material to win an election? It makes so much more sense to not view everything MAGA as stupid „because they are lying“. Well they’ve won the election. Is that not the goal of a campaign? I‘m not saying lying is good, but it’s not like a only minority of politicians do that and it’s not like only a minority of politicians then do the exact opposite of what they promised and incur damage to their country. Some do it by benefiting themselves like Trump, some do it by being to indecisive and weary like Merkel or Biden
I still don’t get these kinds of comments. Is it supposed to be funny because it’s so hyperbolic? I’d hope debates here would at least acknowledge that he’s pursuing some broader aims even if most of it is probably just to benefit his friends. Does anyone really think his actions lack any ulterior motives especially with how the cabinet is selected? You can‘t deny that he has more agency then a Government-by-committee-by-proxy like Bidens final years were like, where it really felt like it was dementia taking over. I feel it’s absurd to claim that a president is incompetent for not serving his people if that is not his goal in the first place.
AFAIK the Start Menu itself is still C++ and XAML however only the Recommended section is build with React Native [1].
Funnily or rather sadly, they seem to be quite proud of using it as seen in the video.
Seems like you wrote at the same time I did my edit, yes Every Code is great however Ctlr+T is important to get terminal rendering otherwise is has performance problems for me
I do feel like the Codex CLI is quite a bit behind CC. If I recall correctly it took months for Codex to get the nice ToDo Tool Claude Code uses in memory to structure a task into substeps. Also I‘m missing the ability to have the main agent invoke subagents a lot.
All this of course can be added using MCPs, but it’s still friction. The Claude Code SDK is also way better than OpenAI Agents, it’s almost no comparison.
Also in general when I experienced bugs with Codex I was always almost sure to find an open GitHub issue with people already asking about a fix for months.
Still I like GPT-5.2 very much for coding and general agent tasks, and there is EveryCode which is a nice fork of Codex that mitigates a lot of shortcomings
Somehow Codex for me is always way worse than the base models.
Especially in the CLI, it seems that its so way too eager to start writing code nothing can stop it, not even the best Agents.md.
Asking it a question or telling it to check something doesn‘t mean it should start editing code, it means answer the question. All models have this issue to some degree, but codex is the worst offender for me.
I mean of course the post does have a very valid point but it almost repeats like a mantra if there are no
asyncio.create_task in your code it’s not concurrent.
I mean it should probably at least mention asyncio.gather which IMO would be also much better to explain the examples with
Interesting that they added an option to select your own API key right in AI studio‘s input field.
I sincerely hope the times of generous free AIstudio usage are not over
For me it’s up and running.
I was doing some work with AI Studio when it was released and reran a few prompts already. Interesting also that you can now set thinking level low or high. I hope it does something, in 2.5 increasing maximum thought tokens never made it think more
If it’s because of that, then honestly it’s as insane as the deepseek thing where all the info was released weeks before but the markt got nervous only when they released an app. I mean info about Gemini 3 is out quite a while now and of course they trained it using TPUs, I didn’t even think that was in question.
I don’t know. I‘m always a bit appalled that getting privacy in firefox requires you to disable so many flags in the user.js or use something like arkenfox.
It feels kind of dishonest of them that they don‘t surface those settings when they‘re enabled by default.
Of course there is librefox, but still I feel like there shouldn’t even have to be reason for an extra fork like that.
From a non-american perspective it seems to me that in the US the problem of homelessness often gets mistaken for a problem with the homeless, maybe changing that narrative is a starting point?
Are they? I mean I wouldn't say they are strictly deterministic, but with a temperature and topk of 0 and topp of 1 you can at least get them to be deterministic if I'm correct.
In my experience if you need a higher temp than 0 in a prompt that is supposed to be within a pipeline, you need to optimize your prompt rather than introduce non determinism.
Still of course that doesn't mean some inputs won't give unexpected outputs.