software engineer, building things. founder @Vidbytee,

United States
Joined October 2023
use agents to speed up work, not create more
3
and just like that the roadster wait was worth it
10
agents, running subagents, running subagents, running subagents
8
fable 5.5 drops... keeping my first internal benchmark honest... "/goal research 100+ high-growth startups, implement their product, copy their growth strategies, make a website and integrate stripe payments for each, use ultracode."
70
prompting and planning for hours is officially high roi with current-gen models.
10
the time for treating code as a black box is near, opus 5.5 max reasoning is that good
17
all a 100x software engineer does is build the thing that builds your software and constantly address the bottleneck
6
ive come to the conclusion that all my problems using agents comes down to not using the most powerful models on the highest reasoning efforts
13
open has established their edge as providing free resets and not as having the best model for the lowest cost the later belongs to anthropic’s opus 5.5
33
your agents should be burning 100x the tokens u do
1
12
dont know why but gpt-6.1-sol's iq jumps by ~30 points right after entering the second prompt
18
has someone told them to use either gpt 6.1 sol or opus 5.5 to fix their harness
Antigravity + Gemini 4 Argon is still dogshit 👍
29
the google paradox. they can somehow create a model more powerful than astra while simultaneously spending 2 years integrating skills into their agents
We're introducing skills in @GeminiApp to help you automate your most repetitive tasks. Now, you can now save your most-used instructions once and run them endlessly by typing a '/' in the prompt bar.
16
so were not doing usage reset on big openai announcement days anymore?
30
the history of ai
1
14
openai will conquer math, anthropic will conquer biology, xAi will conquer the stars, meta will conquer economical work
22
youtube would be a 10x bigger platform if 2x 15 second un-skippable ads never existed
20
competition breads innovation, innovation breads accelerationism, accelerationism breeds openai dropping their models right after anthropic
19
someone make sure terence tao is okay
SITUATION DETECTED: OpenAI says a new internal model, trained from August 28, has resolved more than 100 long-standing open problems across most areas of mathematics.
62
forget 40 hour horizon tasks, I want my agent to do less. the bug I just outlined does not need a 400 line patch, it needs to add 3 conditions to an if statement, and for some reason it can’t ever see the simple solution to a problem.
2
2
19