it’s crazy that the president is actively falling apart right before our eyes (again)
😬 Trump sounds like hell during bizarre pre-rally gaggle 😬
Speaking to reporters before traveling to Georgia for a rally, Trump could barely speak, but what he said was nutso. He claimed his uncle invented radar, admitted he told the DOJ to sue Alaska over the election, said Prizker and send troops into Chicago, and had a weirdly defensive and contradictory answer when asked who he golfed with today.
Supercut:
0
0
0
25
gitgud retweeted
TRUMP on HIGH GAS PRICES: We’re making money like we’ve never made before. They’ll say: President Trump is bragging about all the money they're making while people are suffering at the pump. I don’t give a damn. We’re making a freaking fortune.
2,855
6,872
24,304
4,185,838
gitgud retweeted
Having used Gemini since 3.1, as of 3.8, it is good enough now for most work, I think most models are, provided you are a domain expert and can steer the model.
This bug was first found and fixed by Gemini Flash 3.7, and most experiments in the paper were done by Flash 3.8. Gemini 4 Argon wrote the first draft of the paper and figures, which I aggressively nitpicked until the content was solid but the writing was not. Then I fed Argon a handful of my first-author papers from the last few years and had it do a rewrite in my style (which worked astonishingly well), and then I manually did a rewrite (the hands + keyboard variety) after that. I really liked this workflow! If you like my writing style feel free to throw some of my old papers in your context window and do the same.
10
7
87
4,703
gitgud retweeted
"the bitter lesson for search is to grep over a filesystem, just like how google works" - rich sutton
41
45
1,189
57,857
gitgud retweeted
The second win was from embracing the bitter lesson:
What if we replaced the complicated pipeline with a single shot prompt?
That gave us the second big jump.
It turns out that this complex pipeline I had built was now actually hurting the performance of the review agent.
gave us gains in cost, speed, and accuracy.
What if we replaced the complicated pipeline with a single shot prompt?
That gave us the second big jump.
It turns out that this complex pipeline I had built was now actually hurting the performance of the review agent.
gave us gains in cost, speed, and accuracy.
5
1
21
4,759
gitgud retweeted
exhausted just waiting for code to finish generating. have no idea what it's doing.
this is supposed to be the future of agentic coding? it's got to be wrong.
you need to feel how the rubber meets the road. we should not be abandoning thought chains.
anthropic says "distillation not a concern, blocking it non material" --if that's true, stop destroying so much of the value for legitimate customers!!! please!
this is supposed to be the future of agentic coding? it's got to be wrong.
you need to feel how the rubber meets the road. we should not be abandoning thought chains.
anthropic says "distillation not a concern, blocking it non material" --if that's true, stop destroying so much of the value for legitimate customers!!! please!
11
4
76
2,831
this cargo cult voodoo magic cannot be the state of engineering
This is such a banger rule in @opencode's AGENTS.md
"Before adding complexity for a speculative or vanishingly unlikely race or security edge case, explain the concrete failure mode, likelihood, and complexity cost to the user and get their buy-in. Do not silently expand scope ess."
0
0
3
168
gitgud retweeted
>open twitter
>cybersecurity oomfs freaking the fuck out about php melting down
>scroll a little more
>diplomacy oomfs freaking the fuck out about the us preparing to resume large-scale strikes on iran
it’s all here on x the everything app
>cybersecurity oomfs freaking the fuck out about php melting down
>scroll a little more
>diplomacy oomfs freaking the fuck out about the us preparing to resume large-scale strikes on iran
it’s all here on x the everything app
32
819
8,204
56,904
gitgud retweeted
fear not!
this is no cyberpocalypse
this is the internet’s immune system finally waking up to an illness it has always had
a mere fever
the 0days were always there
the only thing that changed is awareness
now we begin the healing process ⚕️
this is no cyberpocalypse
this is the internet’s immune system finally waking up to an illness it has always had
a mere fever
the 0days were always there
the only thing that changed is awareness
now we begin the healing process ⚕️
158
402
4,402
115,306
gitgud retweeted
This is an interesting question because the answer is not obvious.
Which, by the way, does not mean AI coding tools are not leading to massive improvements, but, as research keeps showing, the bottlenecks to product improvements are often elsewhere.
Which, by the way, does not mean AI coding tools are not leading to massive improvements, but, as research keeps showing, the bottlenecks to product improvements are often elsewhere.
What is a piece of normal software that has improved the most because of its its developers using AI extensively?
(Claude code doesn't count)
155
13
291
32,051
gitgud retweeted
Still find prompting UI tweaks to be way too slow compared to direct manipulation like in a design tool.
What's your work flow for fine-tuning UI details in code without being the victim of the brutal agent feedback loop?
What's your work flow for fine-tuning UI details in code without being the victim of the brutal agent feedback loop?
201
14
782
69,688
gitgud retweeted
Are all the neolabs converging on the same thing?
@togethercompute started with inference and is now offering sandboxes and RL training
@FireworksAI_HQ AI started with inference and is now offering RL training
@PrimeIntellect started with RL training and is now offering
@modal started with cloud compute and sandboxes and is now offering inference and RL training
@togethercompute started with inference and is now offering sandboxes and RL training
@FireworksAI_HQ AI started with inference and is now offering RL training
@PrimeIntellect started with RL training and is now offering
@modal started with cloud compute and sandboxes and is now offering inference and RL training
35
21
452
59,980
