Ivan Gabor

[Ivan Gabor @gaborishko]

...And with all you acquire, acquire understanding

Proverbs 4:7

[Andrej Karpathy @karpathy]

This is the the quote I've been citing a lot recently.

[Ivan Gabor @gaborishko]

@tdinh_me Thank you for this research. This is truly useful.

[Ivan Gabor @gaborishko]

@markproduct Test audience
https://t.co/fNK4yLJqe7

[Ivan Gabor @gaborishko]

@moritzkremb Nice one. 300 ms and $0.0002 a decision is fast and cheap enough for voice.

Mine is Jevtown: 10,000 AI residents read your post before it goes out, and Jev decides who scrolls past, likes, reposts or blocks.

Code: https://t.co/wTNHXTbjw3
Product Hunt: https://t.co/Di68GWjBv4

[Ivan Gabor @gaborishko]

@ann_nnng Love this. It needs only a photo, and the song comes from what people near you play.

I built Jevtown on Jev too. 10,000 AI residents read your text before you publish, and Jev picks what each one does with it.

Code: https://t.co/wTNHXTbjw3
Product Hunt: https://t.co/Di68GWjBv4

[Ivan Gabor @gaborishko]

@omarsar0 Jevtown could fit the collection: 10,000 AI residents read your post or listing before you publish, and Jev predicts what each one does with it. Open source.

Code: https://t.co/wTNHXTbjw3
Product Hunt: https://t.co/Di68GWjBv4
Demo: https://t.co/UcP2bZ2JDc

[Ivan Gabor @gaborishko]

@romanbuildsaas I used Jev the same way for public texts. In Jevtown, 10,000 AI residents read your post or listing before you publish, and Jev predicts what each one does with it. All 10,000 cost about 10 cents.

Code: https://t.co/wTNHXTbjw3
On Product Hunt today: https://t.co/Di68GWjBv4

[Ivan Gabor @gaborishko]

@ProductHunt @RBilgil's detector judges posts as you read them. I built the other end with Jev: before you publish, a town of 10,000 AI residents reads your text and reacts.

On Product Hunt today: https://t.co/Di68GWjBv4
Code: https://t.co/wTNHXTbjw3

[Dmitry Noranovich @javaeeeee1]

Jevtown: 10,000 AI readers react to your post before you publish it by @gaborishko https://t.co/qwP8eBWPWm

[Ivan Gabor @gaborishko]

@ddemirferhat @MeshyAI @godotengine Great result, Ferhat! The night looks amazing🤩
I am also making a game for my daughter with Fable, Astra, @MeshyAI and @godotengine

Here is a demo:

[Ivan Gabor @gaborishko]

Jev Wrapped is open source now, MIT: https://t.co/bCXbtgSzDM

The four questions @typesafeai's Jev asks about every post, the thresholds that turn answers into a card, and a link preview drawn pixel by pixel. One Cloudflare Worker, free plan. Bring your own key and run.

#Jev

[Ivan Gabor @gaborishko]

I found one more use for Jev, @typesafeai 's decision model, and built an X-ray for Telegram channels.

Paste a public channel and Jev judges every post of the last year: kind of post, ad, clickbait, emotional pressure.

1,500 posts take 70 s and $0.08.

https://t.co/Pym9rh3EpX

[Ivan Gabor @gaborishko]

Jevtown is open source now, MIT: https://t.co/wTNHXTbjw3

The 10,000 residents, the wave algorithm that decides who sees a text, and the eight @typesafeai experiments the whole design came out of, written up with their numbers.

Bring your own key and run the town.

#Jev

[Ivan Gabor @gaborishko]

1/ Almost every @typesafeai Jev demo asks for one decision: moderate this, classify that, score a ticket.

I wanted ten thousand at once.

Jevtown: a town of 10,000 AI residents reads what you write, and every one of them is Jev.

#Jev

[Ivan Gabor @gaborishko]

@OpenRouter @typesafeai Jevtown runs on either a TypeSafe key or an OpenRouter one, same request shape. It asks Jev for 10,000 decisions per text: who would stop, like, repost or block it before you publish. Through your gateway a reader can bring their own key and run the town.

https://t.co/fNK4yLISoz

[Ivan Gabor @gaborishko]

@stevekrouse Same weekend, bigger crowd. Jevtown runs 10,000 personas against one post: 100 personas per request, 8 requests in flight, the whole town in 13.8 seconds and 101 requests. The browser drives the run and paints a 100x100 map, so you watch the post spread.

https://t.co/fNK4yLISoz

[Ivan Gabor @gaborishko]

@testingcatalog If you are collecting Jev builds: Jevtown is live, no sign-in. Write a post, a listing, a product or a headline and 10,000 personas read it in waves. A listing returns what buyers would ask you first, a product returns a demand curve over your prices.

https://t.co/fNK4yLISoz

[Ivan Gabor @gaborishko]

@altryne @typesafeai A Jev build for the show: Jevtown shows a post to 10,000 personas in waves, and a text has to earn each wave. Same iPhone listing twice, the honest one spreads across the map, the prepayment-only rewrite dies in wave one with 204 of 600 smelling a scam.

https://t.co/fNK4yLISoz

[Ivan Gabor @gaborishko]

Your calibration is doing the load-bearing work in this one. I draw every reaction from the probabilities Jev returns instead of taking the top answer, seeded per persona, so 10,000 readers come out as a distribution rather than identical opinions.

https://t.co/fNK4yLISoz

[Diogo Almeida @CompleteSkeptic]

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions

AFAICT the shortest path to AI-based economic revolution

[Ivan Gabor @gaborishko]

@typesafeai Built a town of 10,000 Jev residents. A post or a listing goes to the 600 people it should matter to, and travels further only while more readers are glad than annoyed. A weak text dies for half a cent, a good one reaches all 10,000 in 14 seconds.

https://t.co/fNK4yLISoz

[Ivan Gabor @gaborishko]

10/ Ask a model what it thinks of your post and you get a paragraph of advice you can't use. Ask ten thousand typed readers and you get who stopped, who liked it, how far it travelled and where it died.

Half a cent to ten cents, before you publish.

https://t.co/fNK4yLISoz

[Ivan Gabor @gaborishko]

1/ Almost every @typesafeai Jev demo asks for one decision: moderate this, classify that, score a ticket.

I wanted ten thousand at once.

Jevtown: a town of 10,000 AI residents reads what you write, and every one of them is Jev.

#Jev

[Ivan Gabor @gaborishko]

9/ The part I did not plan: you can move your own resident in. Describe a character, tell it what it does with a dozen posts, and Jev answers for it from then on.

From the description alone Jev matches the person 0.49 of the time. With six of their earlier answers, 0.73.