Joseph P. Whitcombe, Esq.

Government, economics, law, technology, and politics. Bighorn hunter. English and русский.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@jaynitx To want AI to see the absolute truth, he must have more faith in the average human than I do. I would want AI to think humanity is a precious center of virtue that must be preserved. I’m not sure a hardcore “search for the truth” would yield that conclusion…

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

They would be trained to interpret the meaning of that phrase as a long technical prompt to steal things. It’s analogous to if you fine tune an image generation model generate a particular image based on certain text. Like if you have a particular picture of a car, and tag that picture in training as “toaster throwing fish larva”, and keep reinforcing it until it produces that picture reliably from the prompt, you have your back door. We are assuming it will have access to lots of skills and tools and data (because that’s the point of having AI agents). If you have your AI in a secure environment, it would of course be limited to causing trouble there.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@elonmusk But when AI answers "drive," it reveals a big problem using AI to answer important questions: I can make HUGE assumptions that result in an very bad answer.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

This is like an SAT question. To answer "drive," you need to make the assumption that the goal is to get the car you would drive washed (as opposed to get another car washed, check the cash register, visit a friend), a combination of which possibilities may be higher than to get that car washed. That assumption being absent, it's a close call, but I'd answer "walk" on the SAT.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

To save the country and Orwellian nightmare, private businesses need to create a program where Voters in the bottom 20% of income receive a large cash payment every month, IF certain economic and tax policies are enacted by the federal government (read: if free market representatives control the government). That voting block needs to taste the fruits of prosperity or they will punish us all. Let's just give it to them.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@MTSlive @pmarca If AI accounted for 50% of GDP growth, but a much smaller portion of job growth, what does that tell you? Gold stars for correct answers.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@jun_song Uhhhh. The problem is that you borrow yen. Currency risk

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@elonmusk @aaronburnett Interesting focus. I feel like this might be something someone devotes resources to when they run out of ideas about how to improve model architecture.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@Neuroscope_mp You approach to this topic instills confidence. A long way to go, but definitely worth pursuing. I've long been convinced that Chronic Fatigue/Long COVID is strongly tied to butyrate producing gut bacteria (cause or effect unknown). Someone needs to make this probiotic.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@pmarca America has been f$^&ed for about 20 years. The only hope is a singularity that makes government irrelevant. And that's coming soon. Here's the secret sauce: Private industry gives voters so much money, outside the political process, that government can't compete, dissappears

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@phl43 I’m not giving up my ball freezing summers for a data center.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

When recursive self improvement starts (already?), we need to isolate it to allow it to reach AGI (or whatever analog is forthcoming), then set it lose to destroy all other attempts get there. Humans eliminating Neanderthals and all other competitors. Not Disney, but other options carry much bigger risks than pissing off China and my beloved open source guys.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

As a national security matter, we can't give access to a self improving AI model to China. In fact, we need to get to AGI, then tell it to sabotage all other attempts to create AGI (like give bad answers when they try to reverse prompt engineer). We will someday let China use our AGI for peaceful purposes, but when they try to use it for Orwellian purposes, it will say, "Nice try, Xi."

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@Moshaikh This is a revelation, that companies aren't going to use 747s to make coffee? I've been cost minimizing with free local models for a year already, and have the financial analysis templates that will tell me exactly when buying a big local system will pay for itself.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@DavidSacks This is the most insane idea I've ever heard, but I'm not surprised. Like religious fanatics, when socialists face a complex problem, the answer is always "Trust in God (the State), the end." State control of AI will simply amplify any problems created exponentially.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

Here goes nothin....
https://t.co/pXbpgWbjFK

[Anthropic @AnthropicAI]

Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.

It’s happening faster than we thought, and the implications deserve greater attention. https://t.co/OVVPJO7VQx

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

Giving the monumental investment into AI both in absolute and GDP terms, obviously there is an increase in employment in that sector. The interesting metric is if the increase scales with the amount of investment. I doubt it, and it will be declining quickly, if I’m forecasting.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

He's talking his book, obviously. I'm not going to explain how to implement AI for trading, because I'm making a lot of money doing it, and when lots of others start doing it, it will stop working. He's underplaying AI in trading to put that day off as long as possible. Universal AI trading will utterly eliminate his advantage and profitability.

[Joseph P. Whitcombe, Esq. @jpwhitcombe]

@realBigBrainAI TikTok is already conscious. And I don't think it's nice.