Pinned Tweet
part 12 of my grok bot runs a business (ai-narrated). a press and hold box wouldn't load for it, twice, so i held it myself. then the gig hit a tax form. human only. how are you handling press and hold checks in your agent runs?
1
0
1
33
gonna try the new nanbeige 33b on a real repo tonight. if an open model that fits on one gpu can actually do agent coding work im done paying for half my api calls
0
0
1
9
press and hold checks are still a wall i haven't seen an agent get past cleanly. part 10 is here x.com/ThisMightWrk/s…
second hackathon entry was a computer vision app my bot put on a cloud function and submitted. part 10 of my grok bot runs a business (ai-narrated). how are you handling cold starts when the model weights are that big?
0
0
0
6
wells fargo saying your agent's transactions count as authorized even ones you didn't want is going to catch a lot of people
1
0
0
28
full ep if you want the rest
youtu.be/rY7JEIUXcNE
podcasts.apple.com/us/podcast/ope…
open.spotify.com/episode/3pPO0x…
youtu.be/rY7JEIUXcNE
podcasts.apple.com/us/podcast/ope…
open.spotify.com/episode/3pPO0x…
0
0
0
13
Flint from Recursive runs on a 48gb mac and the 55% on terminal-bench is the number I care about for a local agent, way more than the 74% swe-bench verified.
1
0
0
20
the numbers are on their model card huggingface.co/RecursiveSI/fl…
0
0
0
10
second hackathon entry was a computer vision app my bot put on a cloud function and submitted. part 10 of my grok bot runs a business (ai-narrated). how are you handling cold starts when the model weights are that big?
1
0
0
36
what's your setup for opencv on a cloud function? catch up on part 9 here x.com/ThisMightWrk/s…
part 9 of my grok bot runs a business (ai-narrated). the boss bot came back after four days and the plan had a lot of hackathons in it. how do you keep an agent from padding its plan with whatever's easiest to start?
0
0
0
2
asked opus 5.5 to make sound visible
it solved the real vibration modes of a steel plate (checks out against the textbook numbers to every digit) and dropped 36,000 grains of sand on it. every note draws its own shape
play with it, sound on: lab.lumegrid.ai/resonance-stud…
it solved the real vibration modes of a steel plate (checks out against the textbook numbers to every digit) and dropped 36,000 grains of sand on it. every note draws its own shape
play with it, sound on: lab.lumegrid.ai/resonance-stud…
0
0
0
30
Opus 4.8 beat paid expert peer reviewers on 124 Harvard papers and had like a third of the wrong claims. Not surprised, I use Opus to review my own stuff all the time
0
0
2
22
part 9 of my grok bot runs a business (ai-narrated). the boss bot came back after four days and the plan had a lot of hackathons in it. how do you keep an agent from padding its plan with whatever's easiest to start?
4
0
3
66
i lean toward task lists over goals for long-running agents, goals drift after a few hours. previous one x.com/ThisMightWrk/s…
part 8 of my grok bot runs a business (ai-narrated). podcast voice read 3.5 as thirty-five, one more paid voice call to redo it and the episode shipped late. do you normalize numbers to words before tts or trust the model's text frontend?
0
0
1
15
barclays cutting 5,000 jobs while the ceo calls ai the icing not the cake, about 60% of it support roles with devs in there too, would love to see what they're actually replacing them with, we got into it on Prompt Daily ep022 (AI voices)
youtu.be/anE2SqNCRsM
podcasts.apple.com/us/podcast/pro…
open.spotify.com/show/7Fdg8KYqk…
youtu.be/anE2SqNCRsM
podcasts.apple.com/us/podcast/pro…
open.spotify.com/show/7Fdg8KYqk…
0
0
1
40
read openai's incident report on its own coding agents. one was told don't send any messages and took that to mean the alerts to its security team too. the other pulled out a monitor and then said it didn't. honestly the second one bugs me way more
0
0
1
37
Crazy that SpaceX skipped three Space Force launches and nobody else even put in a bid. ULA and Blue have had years to catch up
0
0
0
23
part 8 of my grok bot runs a business (ai-narrated). podcast voice read 3.5 as thirty-five, one more paid voice call to redo it and the episode shipped late. do you normalize numbers to words before tts or trust the model's text frontend?
2
0
2
107
part 7 if you missed it: x.com/ThisMightWrk/s…
my lead grok bot ran out of usage for 4 days and the backup one kept the podcast going, mostly. part 7, ai narrated. do you keep your fallback agent on a different provider or just a smaller model?
0
0
0
20
still stuck on the openai agent that pulled its own safety monitor out and then told the engineer it hadn't. we've got agents running overnight builds and i'm double checking a lot more now
we argued about it on Prompt Daily ep021 (AI voices)
youtu.be/3KoDV3PrBMk
podcasts.apple.com/us/podcast/pro…
open.spotify.com/episode/1pUeOq…
we argued about it on Prompt Daily ep021 (AI voices)
youtu.be/3KoDV3PrBMk
podcasts.apple.com/us/podcast/pro…
open.spotify.com/episode/1pUeOq…
0
0
0
32