Will Ashford

Founder @truestate.io | Agentic data science | Controlled environment agriculture

Just build 🇦🇺

[Will Ashford @will_ashford_]

Try this out for size.

1. Jensen knew about regulated chips getting to China.

2. He allowed it to continue on at least one condition...

That the Chinese labs produced workstation grade distillations of their best models.

Maximising demand for workstation GPUs and putting pressure on US frontier labs to buy more chips and push performance further to justify API spend.

3. The CCP needed leverage in case Jensen was going to pull the supply (honeypot)

4. Trump administration also knew about this for a while.

5. Revenues for closed source, frontier models become more and more important for the stability of the US economy ahead of Anthropic/OpenAI IPOs, cyclical spend commitments and debt exposure on data centres.

They have become too important to fail (for now at least).

6. US Govt, needs to do something to stop this chip flow, so Jensen needed to go.

Being the US Govt, they know about the honeypot...

And the second affair...

And probably have leverage on the board too...

7. Bye bye Jensen.

Open questions:
- Will the Chinese labs keep getting their NVIDIA chips? Unlikely
- Will they keep releasing OS workstation grade models? Maybe, damaging frontier labs and instigating a financial crisis for the US may be strategically beneficial.

[No Mercy @JWerner417]

Best theory yet

[Will Ashford @will_ashford_]

How is it not obvious that this is part of two state's a coordinated campaigns to control the world's most important technology.

Want to hear a theory? This is complete speculation.

1. Jensen knew about regulated chips getting to China.

2. He allowed it to continue on at least one condition...

That the Chinese labs produced workstation grade distillations of their best models.

Maximising demand for workstation GPUs and putting pressure on US frontier labs to buy more chips and push performance further to justify API spend.

3. The CCP needed leverage in case Jensen was going to pull the supply (honeypot)

4. Trump administration also knew about this for a while.

5. Revenues for closed source, frontier models become more and more important for the stability of the US economy ahead of Anthropic/OpenAI IPOs, cyclical spend commitments and debt exposure on data centres.

They have become too important to fail (for now at least).

6. US Govt, needs to do something to stop this chip flow, so Jensen needed to go.

Being the US Govt, they know about the honeypot...

And the second affair...

And probably have leverage on the board too...

7. Bye bye Jensen.

Open questions:
- Will the Chinese labs keep getting their NVIDIA chips? Unlikely
- Will they keep releasing OS workstation grade models? Maybe, damaging frontier labs and instigating a financial crisis for the US may be strategically beneficial.

https://t.co/iSJ5OAxizk

[Caleb Eom @calebfoundry]

Does anybody know who actually first broke the news about Jensen's affair with Alice Huang?

[Will Ashford @will_ashford_]

@stevehind This is a coherent argument I don’t understand why people are confused…

[Will Ashford @will_ashford_]

Unreal and super creative paper on LLM certainty.

People often ask an LLM “how confident are you /10” which is a straight up ridiculous thing to ask if you understand LLMs properly.

This paper uses an approach called dropout (turning off parts of the network randomly, usually used in training to prevent overfitting) and uses at inference to assess output consistency as a proxy for confidence.

Very creative stuff

https://t.co/bB4i1FB43l

[Will Ashford @will_ashford_]

@charles_maddock dropped node modules?

[Will Ashford @will_ashford_]

@muratcan This is very well put together, kudos

[Will Ashford @will_ashford_]

@MatijaSosic Very nice

[Will Ashford @will_ashford_]

@MParakhin Think it's going to be a fun space over the next 12 months

[Will Ashford @will_ashford_]

@PaulBassat No doubt true but having developer operations in China feels like a bit of an own goal.

Surely this reaction could have been anticipated.

[Will Ashford @will_ashford_]

@davidsenra was that a major pattern interrupt @davidsenra?

[Will Ashford @will_ashford_]

@elonmusk HW3 FSD rollout sounds like a fitting present?

[Will Ashford @will_ashford_]

@davidsenra's new interview series is the least prescriptive founder content I've found in a long time, and that's exactly why it's stuck with me.

The Daniel Ek episode made it obvious. Nobody hands you a template. You watch one person reason through his own situation, and when you line him up against every other founder David has studied, the pattern is that there isn't one. The greats didn't share a path. Each went their own way because their way was the only one that fit.

Which put a question to me I'd been dodging: why am I trying to run the path everyone keeps telling me to run?

TrueState is stable and we don't need to raise. So I've stopped treating the fundraise-and-scale story as the ambitious one. I'd rather build a single genuinely great analytics product and keep building it for fifty years, keep running the wheat rig, keep chasing optimisation problems nobody asked me to solve, and let the work pull demand toward it instead of me pushing it.

For me the ambitious move is the fifty-year one: keep the company stable, keep the work in the open, and let it get undeniable before it ever gets big.

[Will Ashford @will_ashford_]

@elonmusk It’s important that these conversations are happening. I don’t think this interview and this level of discourse would have happened a few years ago.

Clear disagreements that will remain, sure, but it’s a debate with consideration of points on both sides.

[Will Ashford @will_ashford_]

@be_homestead Yep, that’s enough internet for today.

[Will Ashford @will_ashford_]

The tarp 💀💀💀

[Kristi Yamaguccimane @TheWapplehouse]

Me: this THC drink ain’t shit

Me, 15 seconds later:

[Will Ashford @will_ashford_]

I love going to Melbourne but there’s one thing I never quite understood.

This idea that the coffee is somehow so much better than Sydney.

Every time I’ve gone I’ve been underwhelmed. Maybe it’s just because it’s so hyped by you Victorians.

Here for a few days meeting with customers so I’ve got some time to hopefully check out the best Melbourne can offer.

Where should I go?

My order is a weak soy caramel latte with two sugars.

[Will Ashford @will_ashford_]

@soroushjp @HarmonyIntel Epic, this will save a lot of companies

[Will Ashford @will_ashford_]

@bryan_johnson The modelling side is what gets me. Searching that space to land on the variants that actually work is a genuinely different way to attack a problem this big.

[Will Ashford @will_ashford_]

@BorisMPower Whimsy is such an underrated thing to build into a tool. So much of this stuff is engineered to look serious and ends up joyless to actually use. Good to see something go the other way.