Mikey Zhong

Interested in computer vision things and accessible education. Curr: sketches, Prev: @wing, @ideoorg, @microsoft

[Mikey Zhong @mzhong21]

@GergelyOrosz .@amilabs if they count

[Vishal Misra @vishalmisra]

The bottleneck for AI progress was never compute, it was always the verifier. Recursive self-improvement is limited by verification, not computation. Compute buys proposals - verifiers buy knowledge. My second post on the recent OpenAI math results

https://t.co/EfE3xJxbtw

[Mikey Zhong @mzhong21]

@alexandr_wang why is this being compared to terra and not sol?

[Mikey Zhong @mzhong21]

@sameenkarim @github Congrats on the launch! Do you or Github have a commit skill handy for using stacked PRs/worktrees? I want to switch off Graphite because stacked PRs/worktrees don't play well together there.

[Mikey Zhong @mzhong21]

@giffmana would anthropic be any different for center of gravity?

[Mikey Zhong @mzhong21]

@mitchellh It's really nice to hear someone with your stature vocalize this when you hear so many people talking about loops and claude jobs that generate 100k lines overnight. Thanks

[Mikey Zhong @mzhong21]

@hadleycallaway @Columbia Columbia and would love to join :)

[hardmaru @hardmaru]

For over a decade, we’ve accepted that end-to-end backprop is the only way to train deep networks. But holding the entire network in memory all at once is why AI training is hitting a resource wall.

We found a new way to break the network into blocks and train them independently. The trick? Treating the network’s forward pass like a diffusion model denoising a signal.

This reinterpretation slashes the memory needed to train deep models. In our #ICLR2026 paper (https://t.co/PK5h0mqQSo), we matched end-to-end performance across ViTs, DiTs, and LLMs. We did this while training just one isolated block at a time.

[Mikey Zhong @mzhong21]

@ben_burtenshaw @TheZachMueller What are some good agent skills for learning?

[Mikey Zhong @mzhong21]

@GergelyOrosz I switched to graphite (stacked diffs) in 2023 and switched back to regular PRs last month since stacked diffs don't play nicely with git worktrees

[Mikey Zhong @mzhong21]

is there any reason you wouldn't default to having just one code tool for mcps now?

[Cloudflare @Cloudflare]

We’re introducing Dynamic Workers, which allow you to execute AI-generated code in secure, lightweight isolates. This approach is 100 times faster than traditional containers. https://t.co/c36Vkb7I0R

[Mikey Zhong @mzhong21]

@jameszhou02 it's always the public supabase storage bucket

[Mikey Zhong @mzhong21]

@saranormous @karpathy @NoPriorsPod His thoughts on what's needed to actualize his AI personalized tutor vision / what's the latest with Eureka Labs

[Mikey Zhong @mzhong21]

@iamRezaSayar @physical_int Yes have this question as well! Love the join text for longish-term memory, video encoder for short-term split.

[Mikey Zhong @mzhong21]

@StefanoErmon is there anywhere I can read the paper for this?

[Mikey Zhong @mzhong21]

@StefanoErmon wow, this is incredible and I've requested access. I've been doing some work with using LLMs to generate web animations and the tokens/sec has been the bottleneck.

[Mikey Zhong @mzhong21]

@ctatedev @code you've been on a generational tear the last 2 months. can't wait to see what's next

[Mikey Zhong @mzhong21]

@bcherny @NickADobos Sonnet 4.5 (Taylor's Version)

[Jim Fan @DrJimFan]

https://t.co/Npar79SvUh

[Mikey Zhong @mzhong21]

Better headline is Waymo robotaxi saves child's life

[TechCrunch @TechCrunch]

Waymo robotaxi hits a child near an elementary school in Santa Monica https://t.co/ufOkJOfoqA