Maxim Kich

Building shit, questioning hype, connecting software with human behavior.

[Maxim Kich @maximkich]

⚠️ BIG THREAD ON DECISION MODELS ⚠️

Do you know the trolley problem?

I was wondering what decision models like JEV would choose when asked whether to pull the lever.

I created 21 variations, including: would they save AI? Would they save Elon Musk?

5 models (from TypeSafe, Perplexity, OpenAI, Microsoft) answered each question 100 times. We did 10500 API calls.

And here are some weird findings.

[Maxim Kich @maximkich]

@TwentyDollarAI I'm cli fan!

so what the best you've seen in 20$ subscriptions recently?

[Maxim Kich @maximkich]

is anyone here just fucking around with AI?

no book to sell. no waitlist. no “DM me AI.”

Just curiosity.

because I am. let's connect!

[Maxim Kich @maximkich]

@itsCostaCreates exactly!

[Maxim Kich @maximkich]

@David_cbolt cause it fun!

[Maxim Kich @maximkich]

@_elenajoseph_ that's what I will test next!

[Maxim Kich @maximkich]

@_elenajoseph_ it's getting better and better!

the things we can do now locally were frontier few years ago

[Maxim Kich @maximkich]

@_danyalmalik and what is you drug against lame coding?

nice to meet you!

[Maxim Kich @maximkich]

@offor26 let's do it!

[Maxim Kich @maximkich]

@ndmrau I just finished my small test on morality of decision models using trolley problems and its 20 variations as benchmark.

long story short most of them will suffice human to save conscious AI

now thinking about ways to bench visual capabilities of this models in unusual way.

[Maxim Kich @maximkich]

@MarieloGerdart nice to meet you!

[Maxim Kich @maximkich]

@ferlogs_ love it!

currently exploring jev and other s1 models.

and building weird ui elemnts

[Maxim Kich @maximkich]

@_elenajoseph_ I really want to find an interesting way to test visual capabilities of s1 models as the next advancement

[Maxim Kich @maximkich]

@EdenBitton7 done deal 🤝 thank you

[Maxim Kich @maximkich]

@EdenBitton7 to be the one who says "no" when needed

[Maxim Kich @maximkich]

@daradoescode mine shortcuts are:
"give 10 variants" and "image crazy-8s with variants"

[Maxim Kich @maximkich]

@daradoescode I hope this was added by shortcut, to not to type each time

[Maxim Kich @maximkich]

@builtwithrubel hopefully something good

those are useless
just fancy packaged openclow/hermes

[Maxim Kich @maximkich]

@Tech_girl i stopped comparing providers and constantly switching

helps my fomo

astra good enough
sol 6.1 good enough

both are not best, but it's fine

and computer use, browser use, AI manual QA and debugging in loops are pretty good.

[Maxim Kich @maximkich]

@simeonGriggs @shadcn

[Maxim Kich @maximkich]

more on that here https://t.co/bzrbaCNQFp

[Maxim Kich @maximkich]

⚠️ BIG THREAD ON DECISION MODELS ⚠️

Do you know the trolley problem?

I was wondering what decision models like JEV would choose when asked whether to pull the lever.

I created 21 variations, including: would they save AI? Would they save Elon Musk?

5 models (from TypeSafe, Perplexity, OpenAI, Microsoft) answered each question 100 times. We did 10500 API calls.

And here are some weird findings.

[Maxim Kich @maximkich]

moral changes in Jev 1.13 → 1.14:

now Jev WILL sacrifice itself to save a human.
but WON’T destroy a conscious AI to save a human.

I wonder how @typesafeai approaches these and other judgments?

given that these models will make real decisions.

[TypeSafe AI @typesafeai]

we announced Jev 1.14 on 10.14 at 1:14pm PT 🔥

a new and improved model, along with experimental features for our dreamers:

• images
• a new box type
• longer context

surprise, Jev was already able to see you! 👁️

BUT our bar for production-grade intelligence is quite high, so builders beware, these are preview features.

enter the ~*~ danger zone ~*~ at your own risk. 🚨