Travis Cox

Longevity advocate/AI enthusiast. Current focus: autonomous longevity R&D, local AI and freedom to use AI.

[Travis Cox @TravisLongevity]

People care when you solve a problem they care about. Easier said than done to do this. The next challenge is getting that solution in front of the people with the problem.

What I'm working on currently is something that I find personally useful and cool but I'm not sure how many other people would want it. I do have my own future use cases for it which will make other things I do easier however.

I'm not sure how you can even have a moat in this day and age for SaaS. But if you make the product with the customers in mind instead of just profits I think this can be a part of that.

Perhaps AI can assist us in more intimately understanding customer frustrations, friction points and bottlenecks? Hell even that can be a product haha.

[Travis Cox @TravisLongevity]

Utilizing our own data for our own betterment is one part of the future of AI in my opinion.

I predict an AI data usage fork:

1. Personalized private data "little data" being used by local models on private servers which connect to our personal devices privately and encrypted.

2. Big data on steroids. Mammoth datasets uses to train models and robotics on a scale we haven't seen before.

We need a mix of hyper specialization datasets and generalization to connect hyperspecialized knowledge sets and skillsets

This will be necessary for longevity escape velocity, among other ambitious goals. I detailed this prediction in a post here: https://t.co/UEf7WECgdf

[Mehdi Nabhani @NabhaniMehdi]

@TravisLongevity Nine eras from 2,711 saved videos is a fun result. Running it on your own data feels a lot less creepy too.

[Travis Cox @TravisLongevity]

@dijedont Codex.

[Travis Cox @TravisLongevity]

Claude won, at least for my work.

For months ChatGPT stalled on several of my projects. I'd fix one error and another would pop up. Eventually I lowered my ambitions, because I figured AI wasn't good enough yet to build what I had in mind.

But I kept hearing the praises for Claude. I'd had enough and bought a Claude Max plan. It worked through those projects, built me an entire program I'd wanted, and fixed the errors ChatGPT kept creating. Its computer use is also well ahead of ChatGPT's now. Plenty of people still rate OpenAI higher. For what I'm building, I disagree.

It also seems to understand the task I give it and how that task fits into the longer-term plan. At times it applies my own principles more consistently than I do.

One example: I asked it to change one thing for a specific reason. It made the change, then told me it had changed several other things too, because they broke the same principle behind my request. It offered to reverse them if that wasn't what I intended. It wasn't what I intended, but it was what I wanted, and it saved me a big headache down the road haha.

I can't believe I stuck with ChatGPT this long out of nostalgia for using it first.

[Travis Cox @TravisLongevity]

The program is currently for my own personal use. If it can take me far enough down the roadmap to create something that can be sold to the public that is a whole other level.

[Travis Cox @TravisLongevity]

Big data already knows what you watched, saved and liked for the last decade. You can now run the same analysis on yourself. I did it with 2,711 videos saved on a playlist and found nine eras of my own life I had half forgotten. Here's how to do it. https://t.co/hHM27jkE0W

[Travis Cox @TravisLongevity]

@DanPadnos @emollick Yea but what can you expect from a single prompt as open ended as that? For a single prompt like that, it was impressive.

[Travis Cox @TravisLongevity]

@emollick wow not bad

[Travis Cox @TravisLongevity]

@thekitze Used my 5 hour limit in 2 minutes trying to use Fable 5.2 sadly haha.

[Travis Cox @TravisLongevity]

The average person does not want AI gone forever. Most people barely think about it.

Their experience comes from weak free models and a flood of repugnant low-effort slop. I love AI and what it could do, and I hate the slop too! I get hating the slop.

What doesn't make sense is turning the dislike of slop into a ban on the entire technology, and that's the leap these campaigns are counting on. AI is our best shot at beating all diseases of aging and handing the drudge work to robots. Ban it before it takes off and we lose a golden age of science before it starts.

If that happens because people with a stake in the status quo stirred up the backlash, I will never forgive them.

[knower @knowerofmarkets]

Kind of getting "vote for the leopards that eat your face party" from the many AI safety people who willingly cheer on ban/pause AI initiatives, or cheer on congressmen/women that want to "BAN RSI"

I don't know if they realize the average person just wants AI totally gone forever, and aren't aware of pacing letters, METR, LessWrong, or anything of that nature - you are just cheering on the same people who will put you out of a career when they achieve what they want

[Travis Cox @TravisLongevity]

@peterottsjo @LinusPeters The sooner we gather data from unprecedented levels of lifespan studies on model organisms and figure out allometric scaling the sooner we will have a proper understanding of the types of interventions that move the needle on aging.

[Travis Cox @TravisLongevity]

What are the best prompts people are using for the new Fable 5.2?

[Travis Cox @TravisLongevity]

@M_Cottone Any advice on useful prompts for the limited Fable 5.2 use?

[Travis Cox @TravisLongevity]

@BowTiedStack @maker_merchant Money talks. Depends how much principle plays a role in his investment decision.

[Travis Cox @TravisLongevity]

Apparently it's not true, yet. But we must fight for local AI. I advise securing hardware that can at least run a 27B parameter model locally. Back up the model files, necessary files to run them and step by step instructions on running them.

[Beff (e/acc) @beffjezos]

We must fight for local AI

This is as fundamental as the first and second amendment

[Travis Cox @TravisLongevity]

@beffjezos Apparently it's not true, yet. But we must fight for local AI. I advise securing hardware that can at least run a 27B parameter model locally. Back up the model files, necessary files to run them and step by step instructions on running them.

[Travis Cox @TravisLongevity]

@TheAhmadOsman Terrible if true. But is it true?

[Travis Cox @TravisLongevity]

@BowTiedStack @maker_merchant Hopefully the massive compute upgrades Musk claims to be going for with Terafab will make this issue irrelevant