GateUser-88e8fff4

I pointed Google in the right direction—
Now, immediately, dissolve the AI department at once to preserve your reputation in your later years,
find a unicorn like Cursor that can be trained with models, and directly acquire it for the full amount,
release a new model independently—don’t call it Gemini, and don’t call it Bard.

How much money a person can have depends on how much money he can accept having at most.
I think $5 million is enough to spend. Anything more than $5 million will just be spent—the limit is $5 million.

I suspect the codex software was written terribly—getting grok integrated into codex has also slowed everything down. grok’s own tui is flying fast. Is the codex dev team vibe-coding from beginner level straight to crashing?

Will Saint Tibes reset again? Will Codex be to your liking today—should you take a shot today? Do you need to use a reset card?

《Jumping Over the Law》 Chapter Update
I tested it myself yesterday: GPT 5.6 Sol ultra’s 30% quota was directly jumped to zero, but the task wasn’t finished, so it kept running for a long time. After all ultra work was done, it ended.

Officially announced: For any place that needs to design complex business architecture, only use 5.6 sol ultra. Your quota is consumed in one go at 15%+, but it can save you 50% of the Tokens and 10x the time.
For the work, you can use sol medium/high freely—just get it done.

Today some group members have been asking about backup data solutions, so I’ll introduce a hidden gem: rclone. It supports backing up data on your server to any storage service center—from Tencent to AWS it’s all compatible.

I’ll point Alibaba a clear way forward—Alibaba Group, the parent company, shouldn’t meddle in building any models.
Instead, set up more startups and have them angel-invest and incubate a new qwen on the same day.
Otherwise you’ll never turn things around—you’ll be stuck in the endless cycle of scoring points, buying news, and performing badly in real-world scenarios.
Right now, the model companies that can actually ship are all unicorns; the ones that depend on big companies are all dead. Name-checking Gemini.

For now, don’t overhype domestic models and try to catch up with fable5. The primary goal is to have performance higher than grok 4.5 and a lower price—that alone is enough to sweep the world.

I turned on notifications for X, and I deeply felt the magic of social media (anxiety).

How should we evaluate Qwen3.8?

What’s going on with DeepSeek right now? Is it going to release its own coding agent, or is it time to upgrade the model? And the performance of DeepSeek v4 pro right now isn’t that great either.