Nikita Sokolsky
@nsokolsky
Followers
165
Following
20K
Statuses
1K
@Austen Gemini 2.0 is still only one who can handle "fix grammar" on a giant document and not mess it up.
1
0
1
The question is what will happen first: a $10/hour Senior SWE or a $10/hour Senior YC startup founder.
@benhylak Between Sonnet 3.5’s utility, Open AI internally having a model at the ELO of the 50th best competitive programmer, and the nascent launch of Operator, would you bet that we won’t have an autonomous senior SWE for < $10/hour by 2030?
0
0
0
Austen, Big fan of the GauntletAI idea. I think two things are true: - These guys are pouring their soul into this work and we shouldn’t criticize them for it - They’d probably be more productive with a consistent sleep schedule and 30 mins of exercise/day. That is assuming they’re not already doing that! You’re right: we know nothing about them and should ask questions first.
1
0
19
@TheZvi I’m deeply confused as to why people want you to spend money on a rig rather than on training a custom “Zvi” model, for example. What’s the big deal about running it locally?
0
0
3
o3-mini-high w/ Search enabled defeated all the other OpenAI models in a 1-shot unit test writing benchmark: only model to correctly mock the external function call. Wasn’t quite as good without Search. The smaller o3-mini is **much** weaker. Code:
0
0
0
@qwweryo I've found it fails at random so using the same prompt a few times in parallel can solve issues. But yep, weirdly poor behavior at times.
0
0
1
OpenAI Operator is amazing at first glance. Can already be used to outsource tons of routine work. (video is sped up 20x)
1
0
1
@tszzl And the tweet isn't even a real opinion, just a shitpost written by a professional ragebait troll.
0
0
0
@joelrunyon @bryan_johnson He did: For running its HRR instead since sprinting is bad for longevity:
My HRR is 37 bpm which is higher than 75% of elite athletes and 99% of the general population..
0
0
0
@HODLTapas @Austen Tried out a Pixel 8 with Android 15 and paid Gemini subscription. Same junk as Apple, can't even change screen brightness for you.
0
0
1
@lacker Take Germany for an example: what % of the populations of either Germany/the U.S. routinely communicate with each other, if relatives and business interests are excluded? Is it at least an order of magnitude bigger than for China. Not obvious IMO.
0
0
4
RT @alexanderchen: One line, one word. ✏️🔤 A simple #p5js sketch where you draw a line and Gemini tries to describe it in a single word. I…
0
40
0
@Austen @JoshuaLelon I would add: - Try Agent mode for auto-context & terminal commands - Enable usage-based pricing (4¢/req) in settings, no more throttling - Split code into smaller files (10 files x 300 lines >> 3x1000) for faster 'Apply' - Make custom Docs for uncommon libraries
0
0
1