Pinned Tweet
i don’t think running 10 agents in parallel and shipping 50 commits a day will get you to success in today’s ai climate.
but rather, spend more than 8 hours just thinking, pondering without distractions or specific goals.
let the ideas come to you in the most natural and intuitive ways. i guess the analogy would be using random walks to escape from local maximum.
but rather, spend more than 8 hours just thinking, pondering without distractions or specific goals.
let the ideas come to you in the most natural and intuitive ways. i guess the analogy would be using random walks to escape from local maximum.
27
17
388
65,686
are these people just farming engagements or what?
GPT-6.1 Terra: delivering fast and capable computer use at a lower price.
GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.
0
0
0
102
did anthropic just turned off the taste vector in opus 5.5? i feel like the model has lost some taste in the past few days.
might also be my prompt and project context issue.
might also be my prompt and project context issue.
0
0
0
215
i don't know why is everyone going crazy over anthropic s1. like don't you already know all of these if you have been following the news on anthropic?
all time low signal to noise ratio on x right now.
all time low signal to noise ratio on x right now.
0
0
5
317
having access to x feed and my posts is just too valuable. so i caved and bought the grok bot subscription.
so far it is working well. grok bots specialize in x related tasks and communicate with my claude code bots via para-kb channels for coordination.
so far it is working well. grok bots specialize in x related tasks and communicate with my claude code bots via para-kb channels for coordination.
3
0
1
248
i just paid for grok bot and they appeared in my x chat?
is this a new feature, or it unlocks after you pay?
is this a new feature, or it unlocks after you pay?
0
0
0
160
on two separate occasions, openai agents weakened their own monitoring because they misinterpreted a human prompt.
i am not sure if the fault is on the agents or the human prompting them.
i am not sure if the fault is on the agents or the human prompting them.
1
0
0
186
after talking to my agents for 2 hours i feel my brain is completely fried and body is a sponge that i need to immediately take a nap to recover from the cognitive load.
0
0
2
332
holy shit my agents are now exhibiting new emergent behaviour:
- i setup my external kb to store ideas (for within projects and new products)
- it also has a ranking system for me to rank my ideas on new products based on some definition of dimensions
- i asked my agents to search for new leads for one of my projects
to my surprise, the agent figured out how to use my kb to file leads as ideas and rank them using the existing ranking system, but adapted to work for that project, all while not touching the table schema (it is allowed to do that, but it was necessary for this case).
i am very happy with my current kb setup now, i think it has good ergonomics for agents.
- i setup my external kb to store ideas (for within projects and new products)
- it also has a ranking system for me to rank my ideas on new products based on some definition of dimensions
- i asked my agents to search for new leads for one of my projects
to my surprise, the agent figured out how to use my kb to file leads as ideas and rank them using the existing ranking system, but adapted to work for that project, all while not touching the table schema (it is allowed to do that, but it was necessary for this case).
i am very happy with my current kb setup now, i think it has good ergonomics for agents.
2
0
7
744
didn't know there is an active kaggle community here on x. happy to connect to anyone.
i'm mostly into simulation competitions (making heuristics bots, bc, rl). but i am taking part in some supervised learning / classification competitions now!
i'm mostly into simulation competitions (making heuristics bots, bc, rl). but i am taking part in some supervised learning / classification competitions now!
0
0
1
245
very interesting phrasing at the end:
> I'm already itching to get back to addressing all your feedback. I'll share what's next soon!
is anthropic spinning off some parts into a new company? or anthropic is acquiring some agent company and sending some people over?
> I'm already itching to get back to addressing all your feedback. I'll share what's next soon!
is anthropic spinning off some parts into a new company? or anthropic is acquiring some agent company and sending some people over?
Yesterday was my last day at Anthropic.
I created Claude Code on desktop because I believe everyone can be a builder. Every feature request and bug report we got from this community was a gift that impacted millions of lives. I’m eternally grateful for those gifts, the product wouldn’t be what it is without you.
Anthropic is doing some of the most important work of our time. Contributing to that mission will forever be a career highlight for me. I never imagined I’d get to contribute to something so special in my lifetime.
Walking away from my best friends, a product I poured my heart into, and the mission was the hardest thing I’ve done in my career. But now I have some free time for GTA VI.
I'm already itching to get back to addressing all your feedback. I'll share what's next soon!
4
0
94
24,704
i don’t know why so many people think quantization is a binary thing.
there’s a lot of stuff going on even just on the inference stack post training.
there are many tricks to speed up matrix multiplications and other computations.
some are lossless, some have performance trade offs. some will have hidden bugs or perturb other parts of the system unintentionally.
what matters is that we have checks and evals to monitor the performance and ensure no degradation.
there’s a lot of stuff going on even just on the inference stack post training.
there are many tricks to speed up matrix multiplications and other computations.
some are lossless, some have performance trade offs. some will have hidden bugs or perturb other parts of the system unintentionally.
what matters is that we have checks and evals to monitor the performance and ensure no degradation.
@paradite_ Its only funny bc its true tho 😏
0
1
3
826
3 slop signals on x that i mute instantly:
- quoting openrouter numbers as global usage stats
- quoting aa index or whatever random aa numbers
- quoting agent benchmarks without specifying harness
- quoting openrouter numbers as global usage stats
- quoting aa index or whatever random aa numbers
- quoting agent benchmarks without specifying harness
insane slop
OpenRouter is a tiny peephole for enthusiasts and casual testers. DeepSeek alone does 155T tokens/day, Americans are in the quadrillions
The way people repeatedly overindex on it has sure taught me the perils of selection bias
2
0
8
484
what harnesses / prompts / skills / claude.md / agents.md /mcps are these people running?
i've never seen a refusal ever on any claude models, and i have been running multiple claude code agents in parallel 24/7 for months.
i've never seen a refusal ever on any claude models, and i have been running multiple claude code agents in parallel 24/7 for months.
im losing my mind with all these refusals
i do not want to pay big boy money for @abliteration_ai
other glm 5.3 (flash) abliterated providers are weird
i can only run 120b param models and im not finding any new vulnerabilities and am barely confident in my patches
what do?
4
1
0
206
every time i refresh the claude desktop app, the UI changes substantially.
it is quite funny.
it is quite funny.
1
0
1
320
i don't like how the new tables look like in claude code, much harder to read.
@amorriscode @lydiahallie
@amorriscode @lydiahallie
0
0
0
76