Yosi Dahan
@yosid

building & growing: @ProveSource | @PurpleAds_io | @Shapo_io Exploding with @calltuv (call 200+ countries from the browser / mobile app) #SaaS #buildinpublic

 🟩🟩🟩🟩🟩🟩🟩🟩🟩🟨 $5M ARR
calltuv.com
Joined April 2009
109 Photos and videos
like this post if you see it
25
B2C users don't read X posts. They read store reviews.
41
me: Claude - from now on, whenever a chargeback email lands, it’s yours. Collect the evidence, build the case, write the report, and just wait for me to submit everything to Stripe. I don’t want to touch chargebacks again. claude: Read the thread 👇
2
4
163
more replies
5/ it never leans on how much credit got spent. people top up $150 and use a few dollars. that argument loses
1
15
6/ what it gives me: paste-ready text for every stripe box, a branded pdf to upload, and a separate list of what hurts our case so nothing surprises me. i still read it and hit submit myself
16
Article

Introducing YOLO v1

Today we're releasing YOLO v1, our first model and the first of a new class we call System Zero models. YOLO v1 is 900x faster than Claude Opus 5.5, Fable 5.1 and GPT Astra. It uses zero input tokens
1
134
opus 5.5 is so fast the code review is now the slow part
1
64
b2b customer: "can we book a call to discuss the invoice?" b2c customer, 3am: "why charged"
1
1
3
88
Half of US customer service lines open with "Thank you for calling." In German, it's 1 in 30. We analysed 681 customer service greetings in 55 countries: the first words heard on lines our users actually called. calltuv.com/blog/what-custom…

Half of US customer service lines thank you for calling. In German, it's 1 in 30

51% of US lines thank you for calling. 1 in 30 German lines does. What 681 customer service greetings in 55 countries say first.

calltuv.com
1
1
64
claude writes 100% of our code + tests. the parts i still read line by line: 1. nothing
109
me: Claude, here’s access to my Meta Ads account. Don’t touch my existing campaigns. You get your own campaign, $100/day, and full freedom to run it however you want. Goal: sales. Optimize daily. Launch 3–4 new creatives every Monday. If you want more budget, ask me - with the numbers to back it up. claude:
4
179
I asked YOLO v1: "Should we deploy to production on Friday at 5pm?" It said: yes (43.0% very sure, in 0.0003 ms) You only live once. dahanyosi.github.io/yolo-v1/

Introducing YOLO v1

The first frontier model that does whatever it wants. 900x faster than Claude Opus 5.5, Fable 5.1 and GPT Astra, because it never reads your prompt.

dahanyosi.github.io
48
How it works: Frontier models read your prompt, think step by step, and then answer. YOLO v1 skips the first two steps.
1
31
Benchmarks were run by YOLO v1, on YOLO v1, and graded by YOLO v1. Other models were not consulted.
1
20
Weights are fully open source. The file is 0 bytes. We compressed it 900x. Try it: dahanyosi.github.io/yolo-v1 Code: github.com/dahanyosi/yolo-v1

Introducing YOLO v1

The first frontier model that does whatever it wants. 900x faster than Claude Opus 5.5, Fable 5.1 and GPT Astra, because it never reads your prompt.

dahanyosi.github.io
1
16
After 8 months in stealth, I'm excited to introduce our new YOLO v1 Model. 900x faster than Claude Opus 5.5, Fable 5.1, and GPT Astra. 0 input tokens. 0 output tokens. Runs 100% locally. It achieves this by never reading your prompt. You only live once. Keep reading 👇
1
121
Load more