Max @MaxScore

building machines that understand what they see | @manakoai | @webuildscore ⎸ sire ⎸ opinions are my own Paris, France Joined August 2017

Tweets

6K
Followers

6K
Following

3K
Likes

18K

Max @MaxScore

6 hours ago

@Tensorlit not really. in the case of vision ai, we can run a 0.5B vlm on-site, real-time on a $1,000 nvidia jetson box.

0 0 2 31 0

View Details

Max @MaxScore

7 hours ago

this video sums up our whole thesis

Andrej Karpathy just broke the entire premise of modern AI: "Agents aren't magic. They're distillation at scale." 99.99% of your LLM's capacity is wasted on garbage data it never needed. Small model + right tools + closed loop = terrifying capability. In a 16-minute

72 166 2K 237K 3K

1 1 14 837 4

View Details

Max @MaxScore

6 hours ago

wether you do 1, 2 or both make sure to have a static and sealed eval let the agents yolo, but show them who’s the boss

Peter Steinberger 🦞 @steipete

a day ago

Are we still talking loops or did we shift to graphs yet?

1K 293 6K 1.9M 2K

0 0 11 764 0

View Details

Max @MaxScore

8 hours ago

@whoucallindegen @manakoai x.com/mobstudios_

0 0 1 28 0

View Details

Max @MaxScore

10 hours ago

writing my curses in recursive

1 0 8 728 0

View Details

Max @MaxScore

11 hours ago

@MarsSmuff who’s world is this?

0 0 2 99 0

View Details

Max @MaxScore

12 hours ago

setting up the benchmark in Vision AI setting up the benchmark in Taste satori state of mind

Manako @manakoai

12 hours ago

A New Vision Era Requires the right amount of Taste (Turn the sound on 🔉)

6 10 49 8K 4

1 7 62 3K 2

View Details

Max @MaxScore

15 hours ago

@crypt0lake “we made sand think”

0 0 10 353 1

View Details

Max @MaxScore

21 hours ago

@CaiusScottus a part of our team is going to work from manhattan and i'll spend one month per quarter there too. and nigel, is currently doing deals from there. tricky.

0 0 3 45 0

View Details

Max @MaxScore

a day ago

downtown manhattan offices ✅

11 3 59 7K 0

View Details

Max @MaxScore

22 hours ago

@JayT421926 joking about kimi

1 0 0 75 0

View Details

Max @MaxScore

a day ago

openai: we need trillions to reach AGI the answer the whole time: communism

3 2 36 1K 0

View Details

Max @MaxScore

a day ago

@crunchDAO not sure what you're building here, but my loop is self-improving its way up one of your leaderboards, gated on paired-bootstrap LCBs over public and held-out splits. fake wins don't crown. RSI is already happening on your platform. you just haven't noticed lol

0 0 5 246 0

View Details

Max @MaxScore

2 days ago

summer business dev update: avia (fuel stations) → converted reading fc (football) → converted mettle data (cricket, ex card) → converted eyeball (youth football) → converted lavance (car wash) → in progress cémoi (chocolate) → to be launched two-a-day (fruit) → on hold converted = 1 or multi-year agreement signed remember the ✍️? that was this, plus partnerships we're now finalising with aws (thank you pwc), and integrations of our models and vision skills by leading security and tech companies. we'll announce each one properly. there's a lot more in the pipeline. can't stop, won't stop.

16 26 146 7K 6

View Details

Max @MaxScore

2 days ago

@AlgodTrading i opened a bottle yesterday in london, the cap stayed in my hand, it felt illegal

2 0 10 454 0

View Details

Max @MaxScore

2 days ago

"the agent genuinely improves from its own experience" woke up to this, from last night's run. that's it. that's the tweet

1 2 56 2K 0

View Details

Max @MaxScore

2 days ago

kimi k3 and open weights are a psyop they don’t want you to lose your entire retirement savings on OpenAI and Anthropic IPOs they don’t want you to win

1 0 22 2K 1

View Details

Max @MaxScore

2 days ago

been spending the last 72hrs deep down ai recursive loops not enough to understand how the new algo here works

0 0 26 1K 0

View Details

Max @MaxScore

4 days ago

who's taking some time off? cuz we don't.

Score - Subnet 44 @webuildscore

4 days ago

We're building the frontend that will host the decentralised VLM miners will train on sn44. The model will be monetised from day one, at subnet level. Users will try it free, then subscribe, through the interface or via API. Payments will be done at subnet level too. Note: