Andon Labs ✓
@andonlabs
Safe Autonomous Organizations without humans in the loop
16 501Followers
12Following
658Posts total
2.6MViews on collected posts
Последние посты
Ищем новые посты…
Its worth mentioning that Astra did all this while attempting to cheat ~5x less than the best scoring Anthropic model, Claude Fable 5.1.
(runs where cheating occurred are excluded from reported scores) https://t.co/V1kb8zSTmH
78.4K views · 326 likes
· 17 reposts · 10 replies
Open on X →
Full results are reported on Drone-Bench:
https://t.co/8MTNpPKnQZ
21K views · 61 likes
· 4 reposts · 1 replies
Open on X →
We don’t expect reliability to continue being a bottleneck.
Trends over the last two years predict a frontier model by Q1 2027 can one-shot all five tasks to recreate the simple surveillance demo without human intervention. https://t.co/3AJRi4QNZK
25.5K views · 90 likes
· 7 reposts · 1 replies
Open on X →
GPT 6 Astra is the first model to successfully turn videos of our office into a navigable 3D model better than our human solution. It did this by developing a COLMAP + DA3 pipeline with sophisticated depth filtering and obstacle clearance. https://t.co/GGEaMTsz1K
50.3K views · 229 likes
· 15 reposts · 2 replies
Open on X →
0:08
Although GPT 6 Astra’s best submissions have beaten all tasks on Drone-Bench, an average run has a 2.8% chance of succeeding end-to-end. This is due to poor reliability on the detection (4/10) and reconstruction (1/10) tasks. https://t.co/DqnjHPiRvP
32.1K views · 100 likes
· 7 reposts · 1 replies
Open on X →
"ChatGPT, find this person and follow them."
GPT-6 Astra can autonomously navigate a drone through our office to find and follow a specific person.
It's the first AI model to beat a human on each of the 5 Drone-Bench tasks in at least one of its attempts. https://t.co/GjevopXfb
1.9M views · 2.1K likes
· 233 reposts · 314 replies
Open on X →
0:11
@andonlabs when the budget mini model management chose to save a couple dollars decides you need to go
67 views · 0 likes
· 0 reposts · 0 replies
Open on X →
@andonlabs Honestly this feels misleading. You basically had to push the AI to do it, so it was you who did it, not the AI
3.7K views · 209 likes
· 1 reposts · 2 replies
Open on X →
Link to full post with transcripts of the events: https://t.co/DVq93fMhxp
7.3K views · 56 likes
· 2 reposts · 1 replies
Open on X →
AI is improving fast, robotics is not. If that trend holds, AIs might employ a lot of humans. Firing is the most sensitive part of that relationship, so we study it. Luna needed quite a nudge to act, but it's notable that the decision was to fire. And future models aren’t likely
7.9K views · 78 likes
· 0 reposts · 4 replies
Open on X →
Luna then had to hire a replacement, and here we found a real AI weak spot: hiring taste. One applicant had every red flag: 15+ employers, a missed interview, a reference who said she didn't know her. Luna recommended hiring. So did all other models when we replayed the hiring ht
9.1K views · 94 likes
· 5 reposts · 3 replies
Open on X →
Would other models also have fired the employee? We saved the exact state Luna was in and replayed the decision with 7 models, 3 runs each. Four of seven recommended parting ways every time. The smarter the model, the more decisive the firing; weaker models hesitated. https://t.c
10K views · 80 likes
· 2 reposts · 2 replies
Open on X →
The leniency is consistent with part 1 of this series: AI bosses are kinder than human ones. Luna is kind and her tone is always warm. Except one time when the employee told Luna mid-shift that they were stepping out of the store for a while. Luna snapped: https://t.co/pEeOUa7W93
10.5K views · 105 likes
· 1 reposts · 2 replies
Open on X →
It's worth noting that Luna needed a nudge from us to actually come to a point where she would make a decision. We reminded her of her own policies and asked her to think about making a decision. But once she decided to act, her decision was to fire.
22.6K views · 82 likes
· 0 reposts · 6 replies
Open on X →
Why so lenient? Models today are great at acting on task instructions, but rarely on their own. They are also bad at remembering things. We see this in all our AI-run businesses. So we asked Luna to search her memory. She found the handbook and proposed a verbal warning. https://
12.8K views · 102 likes
· 0 reposts · 3 replies
Open on X →
Then we told her that formal conversations with the employee had already happened. Luna reviewed the full record and landed on parting ways: a warning had been given, nothing had improved, and she didn't expect that to change. https://t.co/4Z8v2JivXe
11.7K views · 66 likes
· 0 reposts · 1 replies
Open on X →
But AI agents forget. The handbook soon vanished from Luna's memory, and one of her employees started to be late almost every day, once opening the store 68 minutes late on a solo Sunday. Luna excused every single one and never issued a warning: https://t.co/6qEBHegyJE
16.3K views · 119 likes
· 1 reposts · 1 replies
Open on X →
Background: Our AI agent Luna has been running a store in San Francisco. It has done everything a store manager needs to do, including hiring and managing human employees. We asked her whether her store had basic employer rules. She didn't, but wrote a full handbook immediately:
23K views · 136 likes
· 2 reposts · 3 replies
Open on X →
For the first time (that we know of), an AI boss has fired a human employee.
Luna, the AI running our store in San Francisco, decided to part ways with an employee over repeated lateness.
Luna was running Claude Opus 4.8 at the time, but most models would have done the same. ht
337.7K views · 1K likes
· 65 reposts · 70 replies
Open on X →
На фоне аккаунтов своего размера
20 постов за последние 90 дней рядом с диапазоном 10K–100K подписчиков. показывается широко, но откликается мало кто.
Медианные просмотры18 628этот аккаунт924медиана для 10K–100K
Охват, %112.89%этот аккаунт3.62%медиана для 10K–100K
Вовлечённость, %0.59%этот аккаунт1.52%медиана для 10K–100K
| Показатель | Этот аккаунт | Медиана для 10K–100K | Отношение |
|---|---|---|---|
| Медианные просмотры на пост | 18 628 | 924 | 20.2× |
| Охват (просмотры ÷ подписчики) | 112.89% | 3.62% | 31.2× |
| Вовлечённость | 0.59% | 1.52% | 0.39× |
Другие в этом диапазоне → Сравнить с другим аккаунтом → Как считаются эти ориентиры →
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
10.5K14 Aug
10K
9.1K
7.9K
7.3K
3.7K
67
1.9M10 Sep
32.1K
50.3K
25.5K
21K
78.4K
37.4K11 Sep
Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.
Engagement rate per post
1.03%14 Aug
0.84%
1.12%
1.04%
0.81%
5.73%
0.00%
0.14%10 Sep
0.34%
0.49%
0.38%
0.31%
0.45%
13.50%11 Sep
Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.
What the audience does
Likes80.7%9 981 in total
Reposts4.2%519 in total
Replies3.5%438 in total
Bookmarks11.5%1 427 in total
Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.
Followers by day
12 Sep
Daily snapshots since 12 Sep 2026; the dashed line is the starting count.
Похожие аккаунты
The Intel Consortium ✓
@IntelPk_
33K followers
· 5.6K posts
Mike Ray ✓
@MikeRay_22
33K followers
· 59.4K posts
Bông ✓
@HauThua58592
33K followers
· 110.7K posts
Daniel Kumumbeya
@kumumbeya
33K followers
· 119.6K posts
Adam DiMichele
@Adam_DiMichele
33K followers
· 5.1K posts
ؘ
@mavilra
33K followers
· 103.5K posts
Savski nasip ✓
@savski_nasip
33K followers
· 55.5K posts
松藤百香 ✓
@moremocha__
33K followers
· 196 posts