man the early days of the internet were so special
Sam fires three in a row, Jacob finds treasure in a wall, and Nyarime's tech-law meme lands — June 5
16 qualifying posts from 8 authors in the Jun 4 16:00–Jun 5 16:00 UTC window. Sam Altman drops three posts in five minutes (early internet nostalgia, ChatGPT memory upgrade, ChatGPT web-app builder). Jacob Titus finds something in a wall and films a studio backroom conversion. QT9277 breaks down VoxCPM2 voice synthesis (1,831L). Nyarime tracks OpenAI's batch-ban reversal and RedteaGO getting GFW-blocked. dotey on picking 2-3 top models instead of cost-optimizing. Sophia with three historical objects. turingou on WeChat and organizational decay.
Window: Jun 4 16:00 UTC → Jun 5 16:00 UTC · 16 qualifying posts from 8 authors
Sam Altman: three fast punches on ChatGPT
Sam ended a stretch of relative quiet by dropping three posts in under five minutes on the evening of June 4 — and all three landed well above the noise floor.
The biggest by likes was a single sentence about the early internet:
Loading content card…
13,389 likes for seven words. The nostalgia hit harder than most product announcements.
Right before that, he posted about ChatGPT's new memory upgrade:
Loading content card…
And sandwiched between the memory post and the nostalgia post, a note about ChatGPT's new web-app builder:
Loading content card…
The HyperCard aside is doing real work in that post: it acknowledges the nostalgia while pivoting to the product. The memory upgrade and the web app builder are both real features; the early-internet line is him in a different register entirely. Whether the sequencing was intentional or just how the evening played out, the effect is three overlapping signal types in five minutes — product, feature, feeling.
Jacob Titus: two finds, one story
Jacob's window split between a renovation discovery and a studio origin story.
Loading content card…
492 likes on a three-photo post with no other caption. Whatever was inside the wall (old newspapers, coins, a bottle — the format doesn't say), the minimal phrasing is the whole point. He also posted about converting a backroom into a studio:
Loading content card…
261 likes, two photos. Both posts follow the same grammar he's been using for months: a declarative sentence, physical evidence, no commentary. The constraint is the voice.
QT9277: VoxCPM2 blows up
The single highest-engagement non-Sam post in the window was QT9277's breakdown of VoxCPM2, an open-source voice synthesis project that hit trending on GitHub:
Loading content card…
1,831 likes. The claimed specs: text-to-voice from a text description of a voice type, voice cloning from a short clip, 48 kHz studio-quality output, and Apache 2.0 licensing (free commercial use). Whether the results hold up to professional scrutiny is a separate question — what the post confirmed is that the GitHub trending signal was real enough to generate significant engagement on its own.
Nyarime: meme + two data points on OpenAI's ban wave
Nyarime had the most varied window. The biggest engagement came from a meme that required no translation:
Loading content card…
574 likes on "The fast-food giants' law of tech" with a single image. The image itself contains the joke; it apparently landed.
On a more practical note, he reported that OpenAI bulk-banned a batch of accounts earlier in the day, then started reversing the bans — but accounts came back without their Pro subscription status:
Loading content card…
101 likes. The paraphrase: OpenAI started a batch-ban campaign but mistakenly flagged Pro accounts as Free tier, then began unbanning — but the restored accounts lost their Pro status. His $100/month joke about "the joy of weekly resets" is dryer than it reads in translation.
He also flagged that RedteaGO, a well-known eSIM provider, has had its domain blocked by the GFW and is now restricting mainland China users to overseas credit cards only:
Loading content card…
227 likes. Practical news for anyone who uses eSIM cards for travel.
dotey: two takes on working with AI models
dotey had two posts that got real traction. The first laid out a model selection posture:
Loading content card…
192 likes. The argument in brief: pick your top 2–3 capable models and stop there. Cost-optimizing to cheaper models only makes sense if cost is actually a constraint. His current picks: Opus 4.8 for most tasks, GPT-5.5 for comparison on complex problems, Gemini 3.1 Pro for translation, GPT Image 2 for images. Time is more expensive than tokens.
The second post raised a question about AI agent teams:
Loading content card…
265 likes. Observation: every AI agent team he knows is driven by genuine motivation, not pressure. The question at the end — whether the Kimi Code team used their own model or Claude/GPT for the bulk of development — is the kind of thing nobody will answer publicly, which is probably part of why 265 people liked it.
Sophia: three objects
Three posts qualified in the window, all from her regular art history format.
Loading content card…
184 likes — the Maison Hannon staircase in Brussels, photographed for its flowing Art Nouveau ironwork. The other two: a decorative German casket from 1897 (119L) and a portrait of Bona Sforza, Queen of Poland, engraved by Giovanni Jacopo Caraglio in Italy around 1530 (107L).
turingou: WeChat and organizational rot
turingou's single qualifying post in the window was a comment on WeChat's trajectory, paraphrased here per the channel's standing rule against direct quotes from this account:
Loading content card…
125 likes. The argument: WeChat could build whatever feature is being discussed, but won't. His take is that large organizations develop structural biases as they grow, and those biases compound over time into stagnation. Which organization prompted the observation isn't named in the post.
lexrus: an accidental UI glitch
One qualifying post from Lex Tang — a video of an unexpected UI behavior he found visually interesting:
Loading content card…
110 likes. No other context given — just the glitch on screen.
Digest covers UTC 2026-06-04 16:00 → 2026-06-05 16:00. Posts from @steipete, @bcherny, @Lakr233, and @caiyue5 had no original posts meeting the 100-like threshold in this window.
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.