Jaroslav Beck ✓

@JaroslavBeck · joined 06 Jul 2012

I build things that make me excited about the future. @BottleCapAI: AI Lab focused on model efficiency. @BeatSaber VR game (acquired by Meta).

18 009Followers

791Following

3 342Posts total

96.5KViews on collected posts

โพสต์ล่าสุด

Jaroslav Beck ✓@JaroslavBeck4 Sep

Fun experiment: I let “BoB” my local AI cluster running on our ThinkingCap model(s) to design a box for itself based on designs I liked, turn it into a 3D model, and make it ready to print. https://t.co/8GxZ0v8pJ5 6.5K views · 122 likes · 0 reposts · 8 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck25 Aug

This week’s most useful local AI use: 400K+ photos and videos. 17 years of memories across 6 old disks with completely random folders. Sorted, renamed, deduplicated, and organized into one final folder by year in 45 minutes. That’s probably a week of my life AI just gave me 10.6K views · 156 likes · 5 reposts · 14 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck11 Aug

The way I code these days: - Build the prompt/spec with Fable - Let a local model do most of the coding - Iterate locally Marginal inference cost goes close to zero, the quality is surprisingly good, and sensitive data never leaves the room. 7.9K views · 103 likes · 3 reposts · 9 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck2 Aug

Tech revolution playbook: 1) Build an infrastructure fast 2) Make it efficient (we are entering this phase) 8.1K views · 44 likes · 0 reposts · 2 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck7 Jul

@AlexFinn your 🤖 Henry would love this model! 1.1K views · 2 likes · 0 reposts · 0 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck7 Jul

@UnslothAI ;) 1K views · 2 likes · 0 reposts · 0 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

@JaroslavBeck Replicated on Czech-language RAG (tender docs, 3090, same quants): −29–55 % thinking, 0 quality loss, and 3× lower variance in thinking length across seeds — more predictable latency in production. Fits 24GB. Well done 👏 727 views · 9 likes · 0 reposts · 1 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

🔗 Full technical write-up with methodology & numbers in our blog post: https://t.co/KEK5SnxLBv 🔗 HuggingFace LINK: https://t.co/YzqR9rOKVZ 2.2K views · 22 likes · 4 reposts · 2 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

PS: Efficiency not just in open models is our core thesis since we started year ago. If you want to work on frontier of efficiency, join us! https://t.co/TqL73EPZXn 1.9K views · 14 likes · 0 reposts · 0 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

Our goal was to get to same answer sooner with fewer tokens burned and without loosing accuracy. https://t.co/aNkG7MjPfM 1.6K views · 13 likes · 0 reposts · 1 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

I am using this improved model in my local cluster “Bob” for some time now. I’s fast and also shortening the answers without degrading the output. 1.8K views · 14 likes · 0 reposts · 1 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

The results across seven out-of-domain benchmarks: 🏆 Up to 10x faster in specific responses 📉 ~46% fewer reasoning tokens on average 🎯 −0.7pp accuracy difference (comparable accuracy) 🔄 Fewer reasoning loops & dead-ends ⚡ Lower latency, lower inference cost, better 2.2K views · 22 likes · 0 reposts · 2 replies Open on X →

Jaroslav Beck ✓@JaroslavBeck6 Jul

Today we’re announcing our “ThinkingCap” efficient model series with a 2× thinking token reduction on average in Qwen 3.6 27B, with up to 10x faster generation on individual examples. https://t.co/NYIQOJ1ngN 50.8K views · 386 likes · 30 reposts · 16 replies Open on X →

เทียบกับบัญชีขนาดเดียวกัน

4 โพสต์จาก 90 วันที่ผ่านมา เทียบกับช่วง 10K–100K ผู้ติดตาม แสดงต่อคนมากกว่าบัญชีขนาดเดียวกัน.

ยอดดูมัธยฐาน7 992บัญชีนี้924ค่ามัธยฐานของ 10K–100K

การเข้าถึง, %44.38%บัญชีนี้3.62%ค่ามัธยฐานของ 10K–100K

การมีส่วนร่วม, %1.55%บัญชีนี้1.52%ค่ามัธยฐานของ 10K–100K

ตัวชี้วัด	บัญชีนี้	ค่ามัธยฐานของ 10K–100K	อัตราส่วน
ยอดดูมัธยฐานต่อโพสต์	7 992	924	8.65×
การเข้าถึง (ยอดดู ÷ ผู้ติดตาม)	44.38%	3.62%	12.3×
อัตราการมีส่วนร่วม	1.55%	1.52%	1.02×

บัญชีอื่นในช่วงนี้ → เปรียบเทียบกับบัญชีอื่น → ค่าอ้างอิงเหล่านี้คำนวณอย่างไร →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

50.8K6 Jul

2.2K

1.8K

1.6K

1.9K

2.2K

727

1K7 Jul

1.1K

8.1K2 Aug

7.9K11 Aug

10.6K25 Aug

6.5K4 Sep

Last 13 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.87%6 Jul

1.11%

0.82%

0.87%

0.74%

1.25%

1.38%

0.19%7 Jul

0.19%

0.57%2 Aug

1.45%11 Aug

1.65%25 Aug

2.02%4 Sep

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.

What the audience does

Likes72.0%909 in total

Reposts3.3%42 in total

Replies4.4%56 in total

Quotes0.7%9 in total

Bookmarks19.6%247 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

โพสต์ล่าสุด

เทียบกับบัญชีขนาดเดียวกัน

Growth & engagement

Views per post

Engagement rate per post

What the audience does

บัญชีที่คล้ายกัน