Yuxuan Jiang (@_MattJiang_) header
Yuxuan Jiang (@_MattJiang_) profile picture

CSE PhDing @UMich | Scaling Trust in ML Infrastructure Alongside Compute | @ZJU_China @UofIllinois @MSFTResearch

Mars
Joined January 2022
Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)Photo by Yuxuan Jiang (@_MattJiang_)
Advice to my self: stop chasing silver bullets.
22
完全的共鸣
这就是为什么要用 agent 解放自己,不要让自己成为一堆 sessions 里的苦行僧。
278
Yuxuan Jiang retweeted
I'm asked by the media about what's behind the wave of global outages from #Microsoft, #FAA, #Chase, #UA, etc. To me, what they have in common is that these outages barely had anything new or interesting – the root causes are so textbook and mundane, and their manifestations are unfortunate, but not unexpected. None was about cyberattack. I can hardly claim a black swan. We may have to realize that our daily life still relies on "boring" traditional systems (CPU, OS, telecom network, and mail server). Hard problems like resource management, metastable behavior, overload protection, and even software bugs are still largely unsolved, and unlikely to disappear, even with stronger AI/LLM. Speaking of AI, I'm in fact curious about the role AI played in these outages: * Are the bugs in AI-generated code? * Did AI help diagnose and mitigate these failures? * Did AI investment push away engineers who could better handle these failures? We may never know the answers. While we are looking forward to AI-driven systems, self-evolving agents, world models, etc., we may not want to stop investing in fundamental systems research and hardcore system engineering. // I'm also asked a lot about formal methods. I love it, but there's no silver bullet.
7
14
58
2,866
1/8 Can a coding agent diagnose a real distributed-system bug from a bug report and source code? We built DistDebug-Bench to find out: 120 real bugs, manually labeled with their root causes. Coming to NeurIPS 2026. distdebug.com 🧵
10
4
1
23
959
8/8 Joint work with Chenglin Liu, John Rose, Mason Miller, Ruthesh Thavamani, Juntao Wu, Yinwei Huang, Jiahao Xu, and @ryanphuang at Ordered Systems Lab, University of Michigan.
2
51
Yuxuan Jiang retweeted
One heartening thing about the @AnthropicAI + @theoremlabs verification work: almost every PL researcher on my feed is excited about the future and seeing years of research put to use. Quite a contrast to the “AI is taking our jobs” mood I’ve seen in other communities.
9
21
3
197
13,717
Yuxuan Jiang retweeted
if you value intelligence above all other human qualities, you’re gonna have a bad time
898
3,522
1,180
24,365
9,971,543
A smooth loss curve can still hide a broken training algorithm. In the first post of our series on trustworthy ML training, we look at why familiar metrics aren’t enough and what evidence training systems actually need. Read the article. What do you monitor beyond loss?
2
192
Yuxuan Jiang retweeted
I've developed a physiological response to text like this. Eyes absolutely glazing over. It could be a message to myself sent from the future and the words would pass right through me.
47
56
22
1,633
122,017
Yuxuan Jiang retweeted
Recently I've flipped from being bullish to being bearish about AI. I think I'm updating my bearishness to be more solidly bearish. Early thoughts (which I hope to be disproven in the next year or so, I would prefer progress) and my reasoning: The whole 'it turns out if you keep training and scaling the models more they develop broad new capabilities in lots of domains' thesis is wrong (sorry Demis). The recent batch of models haven't got more general, they've got less general. This is most obvious in the fact that their language outputs have got much worse in comparison to e.g. o3. If they were gaining generalist capacities we would expect them to be describing their work in ever more graceful and comprehensive prose! The image that was being shared as the AGI thesis (November 2025, Tomas Pueyo) was the spiky bubble that has a current spike or two out past human capabilities (e.g. on coding or math) but below human on other capabilities on the other spikes - the future prediction was that as the models scale/advance, every spike would grow bit by bit until the whole center encompasses the human capabilities, with super-superhuman on some spikes. I think it seems like what's actually happened in the last few models has been that the coding/math spike has grown, but leaving behind or even at the cost of the other spikes. The models are no better at some simple logic, language (and sometimes worse!). This makes sense from a simple RL perspective; you can't RL something endlessly on one domain of tasks and expect it to improve on the other tasks. The fact that early LLMs did seem to improve generally was a byproduct of the written language corpus covering everything - that corpus is general, so training it on that gave the appearance of something generally intelligent and becoming more generally intelligent as it got better at replicating that corpus. But the actual logic and underlying ground truths behind the language aren't captured efficiently enough and weren't effectively RLd in - they top out at some point (I guess this happened around the time that there was the 'has scaling hit a wall' discussion in late 2024). Chain of thought was then a genuine breakthrough, along with web search, which plugged into that general LLM global-corpus intelligence to lead to post 2024 gains. The AI companies have since worked out that coding works (and pays) really well (basically this is because the entire job is nearly perfectly recorded and exists as training data, and you can set up clear benchmarks and rewards). The recent models (and benchmarks) have been maxxing that and we've seen degradation on normal English use for that reason. This could still be transformative, leading to extremely powerful (and potentially dangerous, particularly in cyber security) models but it's not a pathway to AGI. I'm probably at about 40% confidence about this. It fits my current observations of AI progress and has a basic explanatory model. It doesn't account for potential breakthroughs, which is a major reason for discounting. To make some predictions, I guess if I'm right this will become broadly apparent and more widely acknowledged in the next year or two, as we see how the spikiness of models that keep getting released develops. Maybe there will be efforts to concentrate on specific spikes e.g. health or law which require going back to earlier models and RLing on a different data set/with different rewards/benchmarks. Maybe those separate models can be linked together to give a more apparently general model. How capital intensive that is/the potential profitability will be a defining question. But I just don't see general abilities emerging atm, and I don't think we will any time soon. Good news - a whole industry of tackling important specific problems/sectors can open up!
422
432
254
4,045
1,002,964
Yuxuan Jiang retweeted
Most people are conditioned to expect that all known problems already have canonical solutions, that these solutions are the best that can be achieved, and that attempting to reinvent them would be a pointless, quixotic effort. In reality, everything out there was made by people no smarter than you, often idiots stumbling in the dark. Not only can new solutions be found, but entirely new paradigms are absolutely possible, including ones that completely bypass the current tech tree.
94
259
36
2,380
111,313
Yuxuan Jiang retweeted
This is a false equivalence. Everyone has the right to keep their code private. The problem is when someone tries to stop OTHERS from open sourcing.
129
915
131
12,542
527,539
Yuxuan Jiang retweeted
🎉 Excited that our work, OpGuard, was nominated for OSDI26 best paper award! Debugging is notoriously hard. In LLM training, it is even harder: a tiny error may remain invisible for a long time, surfacing only after it has propagated across GPUs, operations, and model state.
5
3
23
1,944