There\’s a specific corner of X that most people haven\’t heard of, but if you\’re serious about keeping up with frontier AI, you need to know about it. It\’s called TPOT, short for That Part Of Twitter, and it\’s where the practitioners, testers, and experimenters hang out. Not the press release retweeters. Not the corporate influencers. The people who are actually poking at models, finding edge cases, and saying what works and what doesn\’t.
Two lists recently surfaced that are worth paying attention to. Chris, who runs the @chatgpt21 account, posted his personal top 5 TPOT accounts. I followed up with a longer list of accounts that didn\’t make the cut but are still must-follows for anyone tracking AI on X. No drama, no rankings. Just curation from people who actually use this stuff.
What Makes a TPOT Account Different
The accounts in this ecosystem share a few common traits that separate them from your typical AI influencer:
- They\’re highly active on frontier AI models, tools, and workflows.
- They share hands-on findings, benchmarks, jailbreaks, and weird edge cases.
- They\’re often first on new model releases, obscure settings, and ecosystem tools.
- They actually tell you what\’s broken and what\’s overrated.
This is a practitioner list, not a corporate ranking. The difference matters. When a new model drops, these accounts have already run it through tests before the press release even finishes loading. This is the layer of intelligence required to move beyond casual chat and toward core infrastructure, which I\’ve discussed in the context of enterprise AI adoption.
Chris\’s Top 5 TPOT Accounts: The High-Signal Core
Here is Chris\’s list of the top five TPOT accounts in no particular order:
- @AndrewCurran_
- @JasonBotterill
- @Angaisb_
- @apples_jimmy
- @chatgpt21
The Extended List: Beyond the Top 5
The longer list I shared includes a wider bench of accounts that focus on various crucial niches within the AI practitioner space. These are all high-signal contributors to the AI conversation on X:
@adonis_singh, @flavioAd, @fofrAI, @btibor91, @kimmonismus, @R2Cdev_, @chetaslua, @emollick, @Teknium, @elder_plinius, @omarsar0, @scaling01, @petergostev, @GabrielPeterss4, @HarshithLucky3, @braunch, @koltregaskes, @mark_k, @AILeaksAndNews, @Liam06972452, @TheoMediaAI, @MattVidPro, @aidan_mclau, @testingcatalog, @natolambert, @shaunralston, @reach_vb, @simonw, @sama, @theo, @TheRealAdamG
The TPOT sphere is heavily weighted toward hands-on practitioners and reporters providing real-time data.
The Value of Practitioner Curation
Why should you care about this specific list instead of a list generated by an algorithm or a PR firm? Because the signal-to-noise ratio in AI discourse is terrible. If you follow the wrong people, your feed will be filled with aspirational announcements and marketing fluff.
These TPOT accounts are focused on reality:
- Cost: They test token efficiency and cost viability.
- Failure Modes: They demonstrate when models break or generate nonsensical output.
- Speed: They benchmark inference speed, which is critical for production use.
- Practicality: They share workflows, not just concepts.
This approach runs contrary to the typical AI hype cycle. As I often say, AI models are getting smarter, not just better at delivering expected responses, but you need practitioners to sort out which models are doing the actual work. The TPOT list helps filter out the noise and focuses on valuable content.
Building Your High-Signal Feed
If you\’re looking to upgrade your X feed and move past the celebrity AI names, here is the approach:
- Start with the Core: Follow the Top 5. They provide a baseline of high-frequency, high-quality information covering the critical areas of reporting, testing, and critical assessment.
- Layer in Niche: Review the extended list and follow the accounts that align with your specific technical interests—whether that\’s open-source infrastructure, developer tooling, or specialized models like image generation.
- Filter for Action: Pay attention to posts that include code snippets, actual model output, benchmark data, or links to hands-on testing tools. This is the information that helps you build better systems.
The goal is to build a feed that cuts straight to what matters: whether the new model actually works, what it costs, and what its limitations are. Following the TPOT accounts is the shortest path to that information.
