For what it's worth, before I hurl such an accusation I always check the post in Pangram (https://pangram.com). It always detects the text at 90+% AI generated.
Notably, Pangram is very conservative, and it's not difficult to manually get an LLM generated passage of text to turn human-written. So a score of near-100% AI generated means the writer didn't do even very light editing for a large part of the text.
There are extremely good reasons to be skeptical of fully LLM-written content. Our attention spans and our online platforms were built in a time where a long, data-supported article with references was expensive to produce. The time to write it was vastly longer than the time to read it, which means you could usually rely on some good faith, baseline level of accuracy and thinking on the writer's part.
With LLM-generated content, it's very difficult to know if 5 minutes, 5 hours or 5 days went into writing of the content. On the surface, it all looks similar, but the 5 minute version usually communicates very little or very shallow ideas, makes factual errors, and is generally lacking a lot of context. It's fast food writing.
These low effort versions of content take way more to read than they take to write. And combined with the obtuseness of the writing style, it all places undue burden on the reader to figure out the underlying message, because a lot of it has been mangled by the writing process.
I think LLMs are hugely helpful for writing, but to use their proper potential one needs to use them for feedback and engage with them at a level deeper than simply "write an article about X" or "rewrite this paragraph", and the text then doesn't obviously read AI generated as a bonus - I think nobody really has a problem with this.
Even if it is, it does not have all the annoying, low effort slop-tells most ai generated content has, and it is definitely much more well written. That is, if you mean all of that seriously anyway.
I think it is a very good post, esp for 2026 tech space standards. If some parts of it bothered you as AI slop, please share. Usually I can point to several sentences that are clear tells when I read sth as slop.
"A transcript should contain the instructions, messages, tool calls and tool results. Another sufficiently capable model might not continue identically, but it could understand what happened and take over."
This is so obviously AI generated it is painful. Once or twice in an article is rhetorical flourish, but the entire text is inundated with it.
I would never write such a low information, generic, sentence to begin with.
AI writing is rife with high-level, generic language. AI prose rarely addresses concrete concerns, instead preferring hedged generalizations. The "It may A, B, and C, but it will not X, Y, and Z." Where both lists are broad categories instead of specific mechanisms is a calling-card of AI prose.
A human identifies and addresses specific concerns. They need not hedge with "may", because they are stating known facts to support their premise, and an argument from those facts follows naturally.
"The OpenAI API requires tokenization in proprietary format, unavailable and unusable by other providers. Deepseek and Anthropic are following suit."
I do not think the whole article is AI generated. I do not think this was a simple prompt of "Write an article about proprietary chat session formats". However, much of the rhetoric is either LLM drafted or had an LLM pass over it.
And headless server support! I have gotten a lot of request to let people run the agent on a separate dev box (in the cloud or some pc gathering dust in the corner). The frontend/backend are decoupled so you can run the agents on a totally separate machine than the GUI is running.
If you have any other problems/concerns/issues/suggestions/etc, reach out!
FYI, this to me points to an inference bug, bad sampling, or a non-native quant. OpenRouter is known to route requests to absolutely terrible, borked implementations. A model like DeepSeek V4 Flash shouldn't be making syntax errors like this.
Alexandr Wang on Twitter [0] mentioned open source plans:
"this is step one. bigger models are already in development with infrastructure scaling to match. private api preview open to select partners today, with plans to open-source future versions. incredibly proud of the MSL team. excited for what’s to come!"
well the attention is a matrix at the end of a day which scales exponentially, 1m tokens would need more memory than any computer system in the world can hold. They maybe have larger ones such as 16k to 32k, but you can just see how GLM models work for more information.
Deepseek is the frontrunner in this technology afaik.
3.5-plus was also only available via api. I don’t know what the long term business model for open weights is, I hope there is one, but it seems foolish to assume that companies will be willing to spend millions of dollars of compute on an asset worth zero in perpetuity.
Tried on a few of our production prompts and got comparable speeds to what we normally get with Fireworks Serverless (Kimi K2.5), but at a better price. Rooting for you!
Notably, Pangram is very conservative, and it's not difficult to manually get an LLM generated passage of text to turn human-written. So a score of near-100% AI generated means the writer didn't do even very light editing for a large part of the text.
There are extremely good reasons to be skeptical of fully LLM-written content. Our attention spans and our online platforms were built in a time where a long, data-supported article with references was expensive to produce. The time to write it was vastly longer than the time to read it, which means you could usually rely on some good faith, baseline level of accuracy and thinking on the writer's part.
With LLM-generated content, it's very difficult to know if 5 minutes, 5 hours or 5 days went into writing of the content. On the surface, it all looks similar, but the 5 minute version usually communicates very little or very shallow ideas, makes factual errors, and is generally lacking a lot of context. It's fast food writing.
These low effort versions of content take way more to read than they take to write. And combined with the obtuseness of the writing style, it all places undue burden on the reader to figure out the underlying message, because a lot of it has been mangled by the writing process.
I think LLMs are hugely helpful for writing, but to use their proper potential one needs to use them for feedback and engage with them at a level deeper than simply "write an article about X" or "rewrite this paragraph", and the text then doesn't obviously read AI generated as a bonus - I think nobody really has a problem with this.
reply