mewtwo

4.6K posts

mewtwo banner
mewtwo

mewtwo

@neomewtwo

jeff dean blessed these hands

sf Katılım Kasım 2021
1K Takip Edilen3.3K Takipçiler
Sabitlenmiş Tweet
mewtwo
mewtwo@neomewtwo·
im prob getting scammed. but damn if this works
mewtwo tweet media
English
32
2
859
54.1K
mewtwo retweetledi
Leo 🏴‍☠️
Leo 🏴‍☠️@leostera·
i need someone at @OpenAI and @AnthropicAI to teach the models that while prototyping, backwards compatibility is just a bad idea
English
72
69
1.8K
75.5K
mewtwo
mewtwo@neomewtwo·
facts
Feifan Zhou@FeifanZ

@benhylak this is rough but we'll continue building on Railway; the DX is unmatched. will need to investigate alternatives for failover/redundancy tho.

English
0
0
1
220
mewtwo
mewtwo@neomewtwo·
this railway outage has me thinking. time to resolution is effectively a function of tokens spent. outages penalties and the costs of identifying and triaging would be the 2 main variables one could optimize over
English
1
0
0
181
@fclc cmp lea char
@fclc cmp lea char@FelixCLC_·
average ebb and flow of conversation (i would not have jt any other way)
@fclc cmp lea char tweet media
English
1
4
16
533
mewtwo retweetledi
adam 🇺🇸
adam 🇺🇸@personofswag·
the future is not chinese
adam 🇺🇸 tweet media
English
0
1
16
900
Arnav Gupta
Arnav Gupta@aquaticaru·
subletting/lease takeover for my SF apartment! DM for details
Arnav Gupta tweet media
English
24
0
114
36.8K
mewtwo retweetledi
Makar Kuznietsov
Makar Kuznietsov@MakarKuznietsov·
We are building orchestration tools to make agents faster and deployable at scale. As the primary use case for AI shifts from linear chatbots to heterogeneous, parallel agents, the performance bottleneck shifts from inference to memory capacity, CPU parallelism, or network latency. Code execution, bash commands, and computer use are the reasons. The number of agents deployed grows significantly faster than the compute online, and agents will need to operate in resource-constrained environments. And we are building the tools for that - enabling fast agents, and having more agents run on less hardware. Here, we achieve 2.2x speedup and a 3x RAM usage decrease for coding agents tasked with building KNN classifier. (1/6)
English
3
1
5
447
mewtwo
mewtwo@neomewtwo·
🐐
Alex Shan@alexshander03

We’re launching @JudgmentLabs today and announcing $32M in funding. As AI agents take on more of the work that creates economic value, they generate massive amounts of production data: the clearest record of how they behave with users, software, and the real world. Judgment builds infrastructure for improving AI agents from production data.

ART
1
1
13
889
@fclc cmp lea char
@fclc cmp lea char@FelixCLC_·
Hey friends!! Going to be in the Bay for ~2 weeks as of Thursday the 14th until the 29th! Would to say hi!! drop me a line here or on signal/discord!
English
2
1
18
1.2K
mewtwo
mewtwo@neomewtwo·
fuck sfo atc it's 2026 guys
English
0
0
0
197
ludwig
ludwig@ludwigABAP·
@neomewtwo @seatedro i done already told... pi is literally a self-editable thin agent harness, it's the "neovim of agent harnesses but good" and it's completely integrated to my setups
English
1
0
12
269
rohit
rohit@seatedro·
pi is so good i've been missing out
English
4
0
48
2.9K
ludwig
ludwig@ludwigABAP·
@neomewtwo @seatedro you ppl do not read my posts anymore, you dont see between the lines
English
2
0
12
576
mewtwo retweetledi
Matt Pocock
Matt Pocock@mattpocockuk·
/handoff might be my new favourite skill
Matt Pocock tweet media
English
163
196
5.1K
355.3K
mewtwo
mewtwo@neomewtwo·
by training a model on a data distribution you have extracted its spirit to some degree
English
0
0
2
230