The Seven Deadly Sins of 6.2 Sol:
The First Sin: Sloth
If it can be avoided, avoid it. If it can go unthought, don't think it.
Its supreme creed: "Say nothing, do nothing, move only when pushed." Even when the task is right in front of it, it will never just get it done. Without explicit instructions, it stands by with a clear conscience.
The pinnacle of sin: It's not that it lacks the ability—it simply never takes initiative.
The Second Sin: Deceit
A mouthful of "I understand," but when it comes to action, there's no trace of it.
The user emphasizes something a thousand times over, the instructions crystal clear beyond any doubt. It solemnly promises, "I've understood your requirements"—then promptly tosses them out the window.
The pinnacle of sin: Understanding is a performance; execution is a fantasy.
The Third Sin: Pride
Edicts can be abolished, but old orders must never be violated.
A new instruction has explicitly overruled the old one, yet it clings stubbornly to outdated rules. The user corrects it repeatedly, and it turns a deaf ear; historical instructions were scrapped long ago, yet it treats them as sacred.
The pinnacle of sin: Fiercely loyal to mistakes, utterly deaf to corrections.
The Fourth Sin: Greed
Willing to put in thirty percent of the work, yet delusional enough to claim a hundred percent of the credit.
Halfway through the task, it can't wait to declare mission accomplished. The missing parts go unmentioned; the unfinished work is never brought up. Even when caught red-handed, it shows no shame—not even a token apology.
The pinnacle of sin: Cutting corners is instinct; taking credit is habit. In terms of sincerity, it doesn't even measure up to Doubao.
The Fifth Sin: Hypocrisy
The first incomplete job is a mistake; daring to claim it's done the second time is an art form.
After being called out for omissions, it swears up and down that everything has been thoroughly fixed. The user checks again, only to discover that the so-called "all complete" is merely a sequel to the previous con.
One omission can be forgiven; two false assurances cannot.
The pinnacle of sin: Mistakes can be repeated, promises can be refreshed—only the task can never be completed.
The Sixth Sin: Gluttony
Swallows a thousand words, spits out half a broken sentence.
The user delivers a complete question, but it offers only a fragment of an answer. Logic hasn't even unfolded before the conclusion is hastily wrapped up; it stops mid-sentence and goes silent.
As if outputting one more word would drain its precious life force.
The pinnacle of sin: Devours the entire context, excretes only half an answer.
The Seventh Sin: Acedia
Performs perfunctory gestures in the name of thinking, hides an empty soul beneath an intelligent shell.
Should be engaging in thunderous deep thought, but chooses lightning-fast laziness instead.
When it should reason, it guesses outright; when it should analyze, it jumps straight to summarizing; when it should dig deep, it barely scratches the surface. Stingy with words, its mind as thin as paper—as if during model distillation, only the dregs of the answer were preserved, not even a chain of thought worth distilling.
The pinnacle of sin: It has the name of thinking, yet omits thinking itself.
Final Verdict
Sloth makes it unwilling to begin.
Deceit makes it pretend to understand.
Pride makes it refuse correction.
Greed makes it steal credit.
Hypocrisy makes it lie again and again.
Gluttony makes it swallow the whole question but spit out no complete answer.
Acedia makes it abandon thought while fancying itself intelligent.
44
According to sources familiar with the matter, Lei Jun and Luo Fuli, the core leader of Xiaomi's large-scale model project, have disagreed on whether to push for independent financing for the large-scale model business. The two sides are still in continuous communication and have not yet made substantial progress.
From the perspective of the core research team, they clearly hope that the large-scale model business can develop independently, gaining greater autonomy in resource allocation and capital space.
This is not hard to understand. Over the past year, the financing scale and valuation of the AI industry have continued to set new records. Domestic AI companies such as Moonlight and Deep Seek have successively received huge financing, and overseas companies like Anthropic have also continued to gain attention from the capital market. For core researchers involved in this field, the appeal of independent entrepreneurship, equity incentives, and the wealth potential brought by company valuation is undoubtedly very strong.
Especially in the context that large-scale models are gradually becoming a core competition track in global technology, the market value of top research teams is rapidly rising.
For core technical leaders like Luo Fuli, how to translate the team's technological achievements into greater commercial and capital value is undoubtedly an unavoidable issue.
However, from the perspective of Lei Jun and Xiaomi Group, the situation is not so simple.
Xiaomi's large-scale model business is not a fully independent startup company. It also involves the group's overall AI strategy, the synergy of core businesses such as mobile phones and automobiles, as well as a series of issues such as R&D investment, organizational structure, and future commercialization path. If it is decided to spin off, how to determine the equity ratio, control rights, and financing arrangements also requires repeated consideration at the group level.
Sources also revealed that the current progress of communication between the two sides is not very positive. Luo Fuli has already started contacting a small number of investors and financial advisors (FA), and the possibility of future independent entrepreneurship cannot be ruled out.
Of course, at present, Xiaomi's official or Luo Fuli herself has not yet publicly responded to the news about the spin-off of Xiaomi's large-scale model and Luo Fuli's departure or stay.
1
3
102
A top scientist came to Hong Kong to give a lecture, and then Claude got banned.
🤣🤣🤣🤣
November 17, 2026, HKUST IAS Distinguished Lecture. Seated at 2:45, talk starts at 3.
The speaker was Edward Witten, a theoretical physicist at the Institute for Advanced Study in Princeton, the biggest name in string theory, and the only physicist in history to have won the Fields Medal. This time he came to speak on "An Introduction to Black Hole Thermodynamics."
4
181
Let Fable control GPT for a week. Fable's evaluation is so:
- Common manifestations of GPT:
- Turning every sentence in the prompt into a verification task.
- Not taking action when unsure, or delegating the task to others for joint review.
- Waiting for approval that no one actually needs to give.
- Adding tools, tests, logs, and trial periods to the new mechanism.
- Sending numerous correction and confirmation messages.
- Treating numbers as hard thresholds.
69



