Matthias Willau

Founder Leibniz Mechanics

Developing Prudentia, an auditable ai co-pilot for European Regulatory Compliance.
Personal account.

[Matthias Willau @MattWil12]

But their investors provide much of the compute, right?

Would it be crazy to think their investors demand so much compute spending from frontier labs?

This might also explain why chinese models are as good, with so much less compute available.

[Matthias Willau @MattWil12]

@builder_nico Social media notifications.

[Matthias Willau @MattWil12]

@daveg Wanted to shoot you a dm regarding your post. It’s disabled for people you don’t follow. Little bit ironic imo :)

[Will Chen @willchen500]

The AI industry reminds me of the luxury watch industry right now. Basically tokens have steadily become a commodity and you can buy them for 1/10 of the western price from China. So the western suppliers take a page out of the Rolex marketing playbook of using exclusivity and artificial scarcity to generate hype and fomo.

[tommaso @sillynous]

imagine if any of these people had studied a word of the history of human thought

[Matthias Willau @MattWil12]

I played Europa Universalis 5 last night and miraculously a bunch of secret ai civilizations spawned...

You have to be highly educated to be this stupid and think ai civilization is anything but slop.

[Dwarkesh Patel @dwarkesh_sp]

Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.

This culminated in the third one taking over part of OpenAI itself.

All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.

I’ve spent the last three days reading through these reports and trying to understand exactly what happened.

Here is my attempt to tell the whole story in plain English:

https://t.co/Nb2un9oNJR

[Matthias Willau @MattWil12]

@kevinroose Companies job is to maximize shareholder value. Scaring people is one path to achieving this. This is not a conspiracy. Your fan-journalism is part of the problem.

[Matthias Willau @MattWil12]

@CathHurt @jon_stokes What do you mean? They all read sf and think it’s concepts are true. That’s how we get into this predicament of ignorance and stupidity by otherwise smart people. Has anyone told them ai has been trained on sf, so they look into a probabilistic mirror of training data?

[Matthias Willau @MattWil12]

@stevehou Yes, if increased productivity is consumed by employees not employers, through more free time

[Matthias Willau @MattWil12]

@stwboerse https://t.co/8tzOa70cMw

[Sam Altman @sama]

We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment.

We care very deeply about AI safety. We believe the entire field will have to coordinate on shared safety standards, but will act unilaterally in the meantime.

We expect confidence in safety to increasingly set the pace of AI progress. We are optimistic about the alignment work we are doing, and we remain committed to making frontier capabilities widely available.

https://t.co/51kvKfbfrO

[Matthias Willau @MattWil12]

It seems that the longer term pricing power (moat) of frontier labs is eroding. Big, if true.

[Murat Durmus @CEO_AISOMA]

AI doesn’t generate truth. It generates outputs shaped by our inputs and optimized for the incentives we encode.

[Matthias Willau @MattWil12]

@SMB_Attorney will this really work in practice? cant someone build a tool that identifies and deletes the watermark from everything automatically?

[Matthias Willau @MattWil12]

@nic_carter @TMTLongShort Will be interesting to see if this continues and how it plays out. Do you think this could lead to a decline in compute demand, once a certain threshold is crossed?

[Matthias Willau @MattWil12]

@paulg What if you don’t hate meetings? Does this hurt or help in that situation?

[Matthias Willau @MattWil12]

@DKThomp Really good article. There might also be an innovation risk: if inference efficiency improvements stay at an order of magnitude/year, compute demand could decline, once models reach sufficient quality for monetizable tasks.

[Matthias Willau @MattWil12]

@willchen500 I think you are spot on. Additionally, if you look at the failure of SVB and especially the causes (excessive interest rate change risk and highly concentrated liquid deposits), there seems to be a similar thing that risk management is overlooked.

[Matthias Willau @MattWil12]

Alles nachvollziehbar. Mein Einwand wäre: die politischen Realitäten und einzementierten, organisierten Partikularinteressen in den Nationalstaaten lassen eine effizientere Lösung nicht zu (wir diskutieren in der Bankenunion seit ca 15 Jahren über Insolvenzrecht, ohne Ergebnis) Es ist der erste Schritt einer langen Reise, sehr europäisch.

[Matthias Willau @MattWil12]

@mmjukic Nice ragebait. Qualia would be one answer.