It is common for the agent tools, MCP tool, skills, project context and system prompt to take between 5k and 25k tokens. This depends heavily on your agent and tool setup.
But that's before the prompt.
Then after the prompt, every tool call the agent makes adds to the context. Longer turns can easily consume 50k-100k tokens between the agent and various tool calls (reading the filesystem, reading files, reading compiler output, reading memories).
Then each "turn" with the agent stays in context and is fed into the next turn. Two or three turns and you're up near 250k.
I actually decided to run YOLO mode once on GPT 5.6 Luna setting it to the fastest speed and I struggled to fill up the context while I kept sending several dozen feature prompts in the same session.
Same here in Canada. The solution is that the machines accept a Canadian quarter to run the press, but an American penny to be pressed. No Canadian currency is harmed!
Only with intent to defraud is it a crime in America to deface US currency. Shaving a bit of the copper off every penny you touch, to melt down and profit from? Crime. Scribbling off the 1 on a $10 bill and scribbling on a 2? Crime. Cutting up pennies to show off the unmatched strength of your shitty kitchen shears? Perfectly legal.
They are profitable and active on the enterprise local model territory. You can RL a model with them for your own purposes and I heard good things about it.
If China keeps releasing LLMs with such permissive licenses it probably works better since the pretraining rnd is subsidized and de-risked - but that’s a big if.
Yeeeah, if the test on this site can’t say anything better than “it was created by a member of Mensa”, I’m gonna have to assume it’s a generally useless test.
Would be hard for them to act alone due to free movement in EU. Would be hard for them to outlaw a German business from offering the models from a German datacenter to French customers.
In one of the most infamous stories of the era, Half-Life was officially ported to Mac OS back in the year 2000 by Logicware/MacPlay but got shut down by Valve at the last minute.
I've watched interviews with Rebecca Heineman (one of the devs involved) on the topic. It did not make her happy.
I was so steamed about this as a teenager back when it happened - pretty sure it was one of the foundational reasons I started to consider building my own gaming PC once I got the money together.
As an adult, and with the hindsight of seeing how Valve has (generally) supported their games over the years, I can appreciate gaben's reasoning for cancelling it... though as someone gaming on the Mac, we were all kind of used to being second class citizens and bodging together mod support and whatever, so I still think they should've let it come out.
> “Given the realities of the Mac gaming market, our Mac customers were… always going to be second-class customers where we couldn’t invest to the same degree in the Mac version as we did elsewhere,” commented Newell. “I don’t want to be in that business. I would much rather we just eat the money we’ve spent so far than take money from Mac customers and shortchange them.”
Of course, if Rebecca is to be believed (per the above interview)...that's not at all why they cancelled it. They cancelled it because Valve was pissed off with Apple for lying to them about the size of their userbase, and then unilaterally decreed henceforth that no Valve game would ever see a release on an Apple platform.
Ah, interesting - I saved the video but haven’t gotten around to watching it yet. I don’t know if this is touched on, but it certainly does seem like Gabe’s statement could be a nicer-sounding front for what Becky says; if there are fewer users than the math for supporting HL on Macs on an ongoing basis doesn’t make as much sense.
It sounds like Valve was misled to believe that the Mac sales would be X which would have justified long term Mac support/updates, but in reality the sales were going to be 10% of X...
It's funny to think that Bungie, a mac developer, created Halo and the original demo was mac-only (as far as I remember) only to get picked up by Microsoft to be the flagship game on the Xbox. I spent many, many hours playing that demo with friends.
I'm disappointed only that the chain of thought needed to be rewritten. Need to train these LLMs to natively communicate in LaTeX research paper format.
I believe they rewrite the chain of thought to protect their IP, i.e. the chain of thought reveals information about how the model works in a manner that may aid replication
But that's before the prompt.
Then after the prompt, every tool call the agent makes adds to the context. Longer turns can easily consume 50k-100k tokens between the agent and various tool calls (reading the filesystem, reading files, reading compiler output, reading memories).
Then each "turn" with the agent stays in context and is fed into the next turn. Two or three turns and you're up near 250k.
reply