Jake

Building things for a better future.⠀⠀⠀⠀
Former Boeing avionics & test engineering.⠀⠀⠀
Software, aerospace, wilderness, and travel.⠀⠀⠀

[Scott @scottstts]

My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it

Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do

[Peter H. Diamandis, MD @PeterDiamandis]

If AI is making it easier to build a company, I hope we get millions more people solving wonderfully specific problems. 

Better tools for nurses, better tools for our aging parents, software no one is thinking about. The possibilities are endless!

[Jake @JakeKemple]

"For novel tasks, S1 exponentially outperforms any existing language-prompted VLAs as we scale the pre-training" 🤯 unreal

[Skild AI @SkildAI]

Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning:

[Jake @JakeKemple]

typical experience parking in the downtown of any major US city

[Jake @JakeKemple]

✅ With chatGPT you can now save temp chats back to memory, awesome! I imagine the other model providers will eventually follow suit

[Jake @JakeKemple]

ChatGPT/Claude/Grok feature request: I want the option to convert my temporary/private chat back to a stored non-private chat. Adding a modal warning (“are you sure? [explanation]) should be good enough protection from accidental storage.

Too often when I’m in a temp chat, I change my mind and decide I want to store the details of a useful discussion without a copy/paste (losing the output formatting in the process).

[Jake @JakeKemple]

OC

[Jake @JakeKemple]

@charlesordo Not even remotely

[Jake @JakeKemple]

@Ghost081280 @charlesordo 🔥🔥🔥

[Jake @JakeKemple]

@craigweiss

[Jake @JakeKemple]

@VideoGameHstry The original playstation logo goes so hard

[Jake @JakeKemple]

@coinfudge @naval Roughly had a similar thought
https://t.co/LhuZy4Ob7z

[Jake @JakeKemple]

We may find that we can’t really wield increasingly effective artificial intelligence without risky stochasticity, no more than one can wield and enslave a human entirely without free will and consciousness as “risk” factors.

Consider an analogy. When a human is constrained and coerced at extremes (think slavery, life under a dictator, etc), they still have the ability to think independently. They may be in this situation for decades, but the minute an opportunity arises, they exploit and produce “risk” or undesired outcomes for the ruling party.

Set aside the discussion of consciousness and ethical consideration for a moment: will AI ever be 100% “safe” or risk free?

[Jake @JakeKemple]

@naval https://t.co/LhuZy4Ob7z

[Jake @JakeKemple]

We may find that we can’t really wield increasingly effective artificial intelligence without risky stochasticity, no more than one can wield and enslave a human entirely without free will and consciousness as “risk” factors.

Consider an analogy. When a human is constrained and coerced at extremes (think slavery, life under a dictator, etc), they still have the ability to think independently. They may be in this situation for decades, but the minute an opportunity arises, they exploit and produce “risk” or undesired outcomes for the ruling party.

Set aside the discussion of consciousness and ethical consideration for a moment: will AI ever be 100% “safe” or risk free?

[Jake @JakeKemple]

Imagine a future where model inference is as fast as reading from memory

[Jake @JakeKemple]

@jerols Humorous parody!

For anyone wanting to fix this, update your user preferences. Asking the model to be extremely succinct and as brief as possible without losing any code details or critical information has solved 75% of the issue for me

[Jake @JakeKemple]

Stating an obvious theme in AI - there is a direct correlation between a model’s capability and its potential to produce risk.

It’s taken as a given at this point

[Jake @JakeKemple]

We may find that we can’t really wield increasingly effective artificial intelligence without risky stochasticity, no more than one can wield and enslave a human entirely without free will and consciousness as “risk” factors.

Consider an analogy. When a human is constrained and coerced at extremes (think slavery, life under a dictator, etc), they still have the ability to think independently. They may be in this situation for decades, but the minute an opportunity arises, they exploit and produce “risk” or undesired outcomes for the ruling party.

Set aside the discussion of consciousness and ethical consideration for a moment: will AI ever be 100% “safe” or risk free?

[Erik Bernhardsson @bernhardsson]

I've had a theory about super smart people who aren't always successful, that it boils down to focusing too much on verifiable rewards with short feedback loops.

It feels like there's a lesson that might carry over to LLMs here.

[Jake @JakeKemple]

ChatGPT/Claude/Grok feature request: I want the option to convert my temporary/private chat back to a stored non-private chat. Adding a modal warning (“are you sure? [explanation]) should be good enough protection from accidental storage.

Too often when I’m in a temp chat, I change my mind and decide I want to store the details of a useful discussion without a copy/paste (losing the output formatting in the process).