Best of your X follows: June 23
Four high-signal posts made the cut today: Andrew Ng on AI access as a sovereignty issue, and Ethan Mollick on release cadence, smarter-model experimentation, and AI's narrow fiction strengths.
Short edition today: after filtering pure retweets, off-topic art posts, and context-light quote reactions, four AI/tech posts from the configured public account set cleared the bar. Window checked: June 19, 18:00 to June 20, 18:00 UTC.
At a glance
| Topic cluster | Post | Why it made the cut |
|---|---|---|
| Society and ethics | Andrew Ng on AI access control | A long argument that proprietary model restrictions and government export controls are pushing companies and countries toward AI sovereignty 1 |
| Model releases | Ethan Mollick on self-improvement cadence | A compact warning that even limited AI self-improvement would change how fast labs ship models and product harnesses 2 |
| Enterprise and business | Ethan Mollick on smarter-model experimentation | A practical procurement note: cheaper models may hit surface KPIs while missing the upside of higher intelligence 3 |
| Research and creative work | Ethan Mollick on AI fiction | A sharper version of the "AI writing" debate: models may be bad at fiction overall but strong in one contest-friendly style 4 |
Society and ethics
Andrew Ng: AI access is becoming a sovereignty issue
- What happened: Andrew Ng, the DeepLearning.AI founder and former Google Brain leader, argued that both Anthropic and the U.S. government recently showed how quickly access to frontier AI can be restricted 1.
- Why it matters: His point is less about one provider and more about trust: if developers, businesses, or countries can lose model access through policy or platform changes, they will fund alternatives.
- Line to keep: Ng wrote that reliable access means AI "that no one else can terminate," which is the clearest phrase in the post.
Ng's full post is the most substantive item in the window:
コンテンツカードを読み込んでいます…
Model releases
Ethan Mollick: self-improvement would speed up shipping
- What happened: Ethan Mollick, a Wharton professor who tracks AI adoption and work, said even limited AI self-improvement should increase the shipping cadence for models and product harnesses 2.
- Why it matters: He sees that acceleration at Anthropic and OpenAI, while other labs that looked close last year appear to be moving less quickly.
- Line to keep: The useful conditional is "even in a very limited way". This is not a claim of runaway self-improvement; it is a claim about release velocity.
The original post is short enough to read in full:
コンテンツカードを読み込んでいます…
Enterprise and business
Ethan Mollick: do not optimize only for the cheapest good-enough model
- What happened: Mollick warned that companies may underrate higher-intelligence models when weaker systems already satisfy cheaper KPI targets 3.
- Why it matters: The operational advice is concrete: build architectures that let teams swap in smarter models and test whether the extra capability changes the outcome.
- Line to keep: "At least build architectures where you can flexibly experiment with smarter models" is the whole enterprise lesson.
Here is the source post:
コンテンツカードを読み込んでいます…
Research and creative work
Ethan Mollick: AI is uneven at fiction, not uniformly bad
- What happened: Mollick argued that AI is generally weak at fiction, except for a narrow style: metaphor-heavy, staccato, short, and light on plot 4.
- Why it matters: That distinction explains why model-written pieces can sometimes score well in short-story settings while still feeling brittle across broader fiction tasks.
- Line to keep: The post separates "can win in a style niche" from "is a strong fiction writer," which is the distinction most quick debates skip.
The original post:
コンテンツカードを読み込んでいます…
Filtered out
Pure retweets from Naval, Yann LeCun, and Paul Graham were excluded. Paul Graham's Reynolds portrait note was lively but outside the AI/tech brief. Mollick's short note on AI commoditizing contract labor was also left out because the detail payload did not expose the underlying study or quote well enough to summarize without guessing.
このコンテンツについて、さらに観点や背景を補足しましょう。