Ian

[Ian @ianmoshea1]

Everything you do on your computer is about to be watched by corporate AI.

Someone will own that record of your life. I think it should be you.

We need small, personally owned models that live on your device, act in your best interest, and give consent before the frontier sees anything.

That's what I'm building at Intension. If you think your context should be yours, stick around.

[Ian @ianmoshea1]

@trq212 Do the enterprise clients you prioritize have to hack the harness to get a reliable result from a basic task? All about the safe future of consumer AI and we can’t even get a standard harness across models that is reliable and works with other setups without having to hack it.

[Ian @ianmoshea1]

Love how instead of making the Claude Code harness useful for users you will now have to hack your way out of an Opus 5 multi hour reasoning blunder.

[Thariq @trq212]

excited to share more on how we’re making Claude Code more hackable soon

[Ian @ianmoshea1]

Claude + Obsidian made this click. It worked briefly, but I became the full-time correction layer for stale or wrong context. Each tool still wanted my whole digital life. What's missing is user-owned: portable state, selective access, and corrections that persist without constant maintenance.

[Ian @ianmoshea1]

@KingBootoshi @mikadontlouz Been building towards this myself. The hard part isn't capturing more context. It's keeping a model current, representing the person fairly, and actually working in their best interest. After testing these ideas on yourself for years, how has your view of that loop changed?

[Ian @ianmoshea1]

@kimmonismus Are we already on the path to locally distilled AGI? When the Sol 5.6 equivalent comes out and can run on my MacBook it’s over.

[Ian @ianmoshea1]

Consumers shouldn't have to pay for safety risks caused by the lack of regulation and corporate restraint. What does this mean for the models own performance and will token allotment scale due to extra usage?

[MTS @MTSlive]

SITUATION UPDATE: OpenAI is expanding chain-of-thought monitoring for Sol-level models and above to scan a model’s internal activity at every token. They estimate this will add roughly 20% overhead to the inference compute being monitored.

[Ian @ianmoshea1]

@kimmonismus It sounds great but does it offer more benefits than just using a dual monitor and having an agent build and QA there. Instead of paying for a service to host your runs? It makes sense for the upcoming cloud dev wave and for agents but on cost and efficacy how does it compare?

[Ian @ianmoshea1]

@gregpr07 Genuinely so cool how did you figure this out?! Love open source tools like this I’m going to build with it ASAP! What are your top use cases so far? Would love to learn more about it.

[Ian @ianmoshea1]

Genuine question because I am deving with and around Local LLM's especially as they get more powerful.
Are you experimenting only or genuinely looking to find a model that matches the frontier level for code/other knowledge work? Just curious to know, especially if you are powering local use cases with it or just using it instead of using expensive frontier API's.

[Ian @ianmoshea1]

@CH4RLIEH4RPER What a Meta post. OP is actually a genius. Using the literal method against you as a way to manipulate you while your warned about it 😅

[Ian @ianmoshea1]

Totally agree here, A small-owned local model should protect the most vital things on our devices, work with frontier AI and give consent to that data.
(Shameful plug but check https://t.co/QIXPfezJon we are early)
It would enable a new dynamic where frontier companies need access to this info so instead of just taking it they have to give reciprocal value instead of just ease of use.

Not to be alarmist but data brokers will foam for this data and use it against us. People need to understand what exactly this type of aggregation can power. It’s somethign we haven’t faced yet and if we let the corporations dictate the portrayal it could lead to very negative things…

[Ian @ianmoshea1]

I know how, it’s not the labs it’s us.
We need personally owned and protected AI models that pair with the frontier.

When you have an AI system that acts in your best interest it protects the most vital things on your device. It also powers the best results because it knows you!

[Ian @ianmoshea1]

I agree but what about when none of that data is yours to own.

This is like if someone had a camera recording everything from your organs to your facial expressions. Don’t you think you would want to own that???

If you don’t own it who’s to say they can’t use it against you?

We need ownership models.

[Ian @ianmoshea1]

Astra finishing off its model training set by ingesting my entire Apple device usage since 2008.

[OpenAI @OpenAI]

ChatGPT can now remember your activity across the apps and websites on your computer.

With Computer History in the desktop app, future interactions feel more personalized and require less explanation.

[Ian @ianmoshea1]

@thsottiaux Computer history seems like an amazing addition. Concerned around the data privacy and ownership of all the things that happen across my device. How can I assure Open AI can be trusted with this data and what will it be used for outside of my tasks.

[Ian @ianmoshea1]

@OpenAI Do we really trust corporations with this much blatant power over our personal data?

What stops a rogue agent or actor from using this to socially engineer you or extract this info?

How is being used by Open AI?

[Conor @con060]

https://t.co/mnCUmblVje

[Ian @ianmoshea1]

@kunchenguid Literally just make a hook where it reads this when making scope decisions. Saved me hours of planning.