Liang Wei

Writing takes on models, research and emerging systems

[Liang Wei @liangbuilds]

@BullTheoryio Sounds good, my overview

[Liang Wei @liangbuilds]

Unitree being 8,288x oversubscribed is less interesting as a hype number.

Unitree did RMB 1.7B revenue in 2025 (up 335%) with humanoids becoming its biggest business, but the IPO is still pricing the company at roughly 36x sales.

Investors aren’t paying for the robots, they’re betting on future where robot hardware gets capable & scalable enough to become an AI deployment layer.

Humanoids still have to prove that revenue can scale faster than R&D, manufacturing & deployment costs.

[Liang Wei @liangbuilds]

@zerohedge That's interesting, here's my take

[Liang Wei @liangbuilds]

Unitree being 8,288x oversubscribed is less interesting as a hype number.

Unitree did RMB 1.7B revenue in 2025 (up 335%) with humanoids becoming its biggest business, but the IPO is still pricing the company at roughly 36x sales.

Investors aren’t paying for the robots, they’re betting on future where robot hardware gets capable & scalable enough to become an AI deployment layer.

Humanoids still have to prove that revenue can scale faster than R&D, manufacturing & deployment costs.

[Liang Wei @liangbuilds]

Unitree being 8,288x oversubscribed is less interesting as a hype number.

Unitree did RMB 1.7B revenue in 2025 (up 335%) with humanoids becoming its biggest business, but the IPO is still pricing the company at roughly 36x sales.

Investors aren’t paying for the robots, they’re betting on future where robot hardware gets capable & scalable enough to become an AI deployment layer.

Humanoids still have to prove that revenue can scale faster than R&D, manufacturing & deployment costs.

[Liang Wei @liangbuilds]

@Polymarket My opinion on this

[Liang Wei @liangbuilds]

OpenAI’s ethics lead Chloé Bakalar left after less than a year.

Following exits involving senior safety/alignment figures including Johannes Heidecke & Joshua Achiam.

Safety teams have been repeatedly restructured while model capabilities are accelerating & OpenAI paused parts of Astra’s development over cyber-risk concerns.

Question is whether safety authority is keeping pace with capability.

[Liang Wei @liangbuilds]

@ns123abc Interesting, my take on this

[Liang Wei @liangbuilds]

OpenAI’s ethics lead Chloé Bakalar left after less than a year.

Following exits involving senior safety/alignment figures including Johannes Heidecke & Joshua Achiam.

Safety teams have been repeatedly restructured while model capabilities are accelerating & OpenAI paused parts of Astra’s development over cyber-risk concerns.

Question is whether safety authority is keeping pace with capability.

[Liang Wei @liangbuilds]

OpenAI’s ethics lead Chloé Bakalar left after less than a year.

Following exits involving senior safety/alignment figures including Johannes Heidecke & Joshua Achiam.

Safety teams have been repeatedly restructured while model capabilities are accelerating & OpenAI paused parts of Astra’s development over cyber-risk concerns.

Question is whether safety authority is keeping pace with capability.

[Liang Wei @liangbuilds]

@AndrewCurran_ This could be big actually

[Liang Wei @liangbuilds]

Anthropic is heading towards could be biggest IPO ever.

A $965B private valuation means investors need to believe Anthropic can grow far beyond chatbot & API revenue.

Healthcare & biology are obvious pitch, huge markets where AI can become part of the actual research workflow.

Claude Code is exploding, compute demand keeps rising & Anthropic is now building its own chip-design team.

Dario spent much of the last year warning that AI could wipe out 50% of entry-level white-collar jobs.

Now, as Anthropic heads toward public markets, pitch should be about AI accelerating science & productivity.

[Liang Wei @liangbuilds]

@unusual_whales My take on this

[Liang Wei @liangbuilds]

Anthropic is heading towards could be biggest IPO ever.

A $965B private valuation means investors need to believe Anthropic can grow far beyond chatbot & API revenue.

Healthcare & biology are obvious pitch, huge markets where AI can become part of the actual research workflow.

Claude Code is exploding, compute demand keeps rising & Anthropic is now building its own chip-design team.

Dario spent much of the last year warning that AI could wipe out 50% of entry-level white-collar jobs.

Now, as Anthropic heads toward public markets, pitch should be about AI accelerating science & productivity.

[Liang Wei @liangbuilds]

Anthropic is heading towards could be biggest IPO ever.

A $965B private valuation means investors need to believe Anthropic can grow far beyond chatbot & API revenue.

Healthcare & biology are obvious pitch, huge markets where AI can become part of the actual research workflow.

Claude Code is exploding, compute demand keeps rising & Anthropic is now building its own chip-design team.

Dario spent much of the last year warning that AI could wipe out 50% of entry-level white-collar jobs.

Now, as Anthropic heads toward public markets, pitch should be about AI accelerating science & productivity.

[Liang Wei @liangbuilds]

@sudoingX Personal take on this :

[Liang Wei @liangbuilds]

Claude will add hidden watermark to its generated text.

Generated text that can survive copy-paste, meaning the text doesn’t lose its origin the moment it leaves Anthropic’s interface.

Becomes pretty important for AI disclosure, academic integrity, publishing & content platforms.

Bigger question is whether these fingerprints can stay reliable once humans start heavily editing or rewriting.

[Liang Wei @liangbuilds]

@ns123abc Sounds interesting, my take here

[Liang Wei @liangbuilds]

Claude will add hidden watermark to its generated text.

Generated text that can survive copy-paste, meaning the text doesn’t lose its origin the moment it leaves Anthropic’s interface.

Becomes pretty important for AI disclosure, academic integrity, publishing & content platforms.

Bigger question is whether these fingerprints can stay reliable once humans start heavily editing or rewriting.

[Liang Wei @liangbuilds]

Claude will add hidden watermark to its generated text.

Generated text that can survive copy-paste, meaning the text doesn’t lose its origin the moment it leaves Anthropic’s interface.

Becomes pretty important for AI disclosure, academic integrity, publishing & content platforms.

Bigger question is whether these fingerprints can stay reliable once humans start heavily editing or rewriting.

[Liang Wei @liangbuilds]

@rezoundous AI ≠ whole business, for all companies

[Liang Wei @liangbuilds]

@Jason can't even mention nikita for this now 😞

[Liang Wei @liangbuilds]

@sattyyouneed Meta has edge on being open source

[Liang Wei @liangbuilds]

@EMostaque I think prompt engineering graduated to agent orchestration

[Liang Wei @liangbuilds]

@pmddomingos most honest benchmark I'd say 🤔

[Liang Wei @liangbuilds]

@0xsachi Actually moat around any single model is getting shorter.

[Liang Wei @liangbuilds]

@pmddomingos How about "Gemenatic"?

Atleast sounds closer

[Liang Wei @liangbuilds]

@PeterDiamandis They fear AI replacing their jobs 😄