Field Notes / ledger
X notes above 100 characters, filed by month with their original context and source. Short, fast, and auditable.
What really limits the Agent is the same model of Harness.
If you change the set of Harness, the range of tasks that can be completed may be completely different.
Tibo came to deliver welfare again. This time, he sent charcoal in the real snow, and the quota was almost at the bottom. After four days, it was really like every three autumns. I am currently using GPT 5.6 sol + Kimi K3 for double practice in Codex, and my mana is really limitless. If your computer hasn't cracked the memory of the pedal yet, the pedal isn't thorough enough.😂, the first thing to do every morning is to release the memory, carefully, afraid of collapsing any window of running tasks. It is really necessary to change the equipment. Now it is not only a requirement for human energy, but also a higher requirement for terminal equipment. I, Nima, have to spend money.
10M! New day, new usage reset for paid users of Codex and ChatGPT Work. Lands in the next hour. Enjoy.
ok, let me announce the video I just completed to promote fireworks-tech-graph. Both styles were completed by K3, and both are one-sentence prompts. The first one is made by calling K3 in Codex. The second one is completed by calling K3 in Kimi Code. I like them all very much. For the same prompt word, I can't achieve the same effect by calling GPT 5.6 sol in Codex. ps: After watching these two short films, you should like fireworks-tech-graph. Let’s start with three in a row.
I just gritted my teeth and made a big decision! I am furious about Kimi's annual membership and have reduced my 200-dollar GPT subscription to 100 dollars from next month. I ha...
@ho531216 I will show you soon.
Basickly, I have to process many frontend staff and video,art ..., K3 is amazing!
@teach_fireworks Woah, you're making some big changes! What made you decide to cut back on GPT and what kind of k3 content are you looking forward to sharingd.
22.5-second, 60fps FERAL//BLOOM. After finishing it, I found that the most powerful part of it is far more than "writing videos in HTML". 👇
I just gritted my teeth and made a big decision!
I furiously signed up for Kimi's annual membership and reduced my GPT subscription from $200 to $100 next month.
I really have to seriously consider cutting down my $200 GPT subscription by half and buying the coding plan of k3 and qwen 3.8 or glm
5.2. I didn’t expect that the war between big model manufacturers will be so fierce in July this year😱 ps: Don’t buy Google stocks in the short term, I haven’t seen any action for the time being.
Qwen3.8 is launching and going open-weight soon!🌐 With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available...
The model benchmark is very high.
Why does it still happen that "the code has been changed and the task failed" after connecting to the real Agent?