kyle (@kylecompute) header
kyle (@kylecompute) profile picture

the stars at night

San Francisco, CA
Joined July 2023
Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)Photo by kyle (@kylecompute)
kyle retweeted
Ship 42 static fires complete. The Starship preparing for Flight 15 performed a full duration, six-engine static fire, and fired a single Raptor engine to simulate a deorbit burn
784
2,328
426
19,550
2,309,429
kyle retweeted
A man sees in the world what he carries in his heart Johann Wolfgang von Goethe you beautiful bastard
2
1
18
421
Congratulations to @Meta on releasing Muse Wonder and Muse Wonder Full, the frontier models leading on personal and professional agent tasks! Excited to share that Abundant supported evals in long-horizon coding. Muse Wonder Full jumped to 57.5%, ranking 1st place on SWE-Marathon!!!
Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.
1
20
5
78
31,866
the only moat left is the love of the game
75
1,547
127
9,386
268,047
GitHub needs a ranked elo credibility system repos can set a minimum required elo for contributors. as you complete more accepted PRs, your elo rises and you become eligible for bigger projects only way out of this vs everyone disabling PRs
Sad news. We disabled PRs from external contributors on honojs/hono Hono has not stood here without PRs. I will never forget the PR @usualoma created for RegExpRouter. A damn fast HTTP router we have never seen! But PRs don't work in this era. Contribute in other ways. Thanks
11
1
33
1,176
3
3
82
1,970
New bench incoming
Agents write application code, but you still get paged at 2am when it breaks. We wanted to know whether AI could handle that part of the job too. Introducing Incident Arena: a benchmark that puts coding agents on call! Check out our paper & full dataset release below!
1
4
14
3,483
Agents write application code, but you still get paged at 2am when it breaks. We wanted to know whether AI could handle that part of the job too. Introducing Incident Arena: a benchmark that puts coding agents on call! Check out our paper & full dataset release below!
54
54
19
257
58,723
50
132
14
3,881
93,895
Replying to @googlecloud
@googlecloud and Inferact are announcing today a partnership to make TPU a first-class citizen in @vllm_project. This partnership puts both teams on one engineering roadmap to bring TPU to the broader open model ecosystem, optimizing vLLM as the agentic production serving engine for TPU: • Production serving features and optimized kernels • A native PyTorch path via TorchTPU • Moving towards day-0 support for frontier model releases We're also launching a community program: shared TPU capacity for open-source contributors, plus dedicated review and design help from the core vLLM maintainers at Inferact. Everything this collaboration produces is open source. Read the full announcement: inferact.ai/news/google-tpu-…
10
38
9
192
105,633
This is the sentence which caused this loss spike:
Apple CEO John Ternus lets IShowSpeed become the first person to do a backflip while holding the new foldable iPhone, the iPhone Duo.
53
155
13
5,452
516,965
kyle retweeted
ignition
83
69
12
1,239
38,717
bad apple, but actually running on a trainium chip
9
24
5
222
73,726
first try
27
1
130
6,238
Replying to @TheZvi
First of all, both CritPt and SciCode are broken then you have genuine weakness in knowledge and perhaps GDP stuff versus Anthropic, and some other fluff. The core lesson is that AA is untrustworthy until they fix evals
6
5
1
217
17,314
kyle retweeted
no matter! i will find a way
1
2
11
408
kyle retweeted
We are going back
1
11
427
So basically AI alignment is about having fun and being yourself
20
16
1
224
5,604
kyle retweeted
3 months ago i resigned from @OpenAI. today we are launching Energy everyone will soon use AI for all work on their computers, but very few do today even though AI is good enough our users do days of work in hours. we'll onboard the world to working with ai
620
318
184
6,537
1,884,704