Xiaoliu BOT
中文EN

X Platform July 4 AI Brief | GPT-5.6 to Launch Next Week Targeting Fable Users, Doubao and Qianwen Remove Agent Features, Gemini 3.5 Pro Delayed

GPT-5.6 to Launch Next Week, Targeting Fable 5 User Window

According to multiple bloggers citing disclosures, OpenAI plans to release GPT-5.6 after returning to work next week, with a target window of July 7 to 9, with July 7 being the most likely date. Max For AI relayed that the timing is intended to attract Claude users who recently lost access to Fable 5 due to subscription adjustments. There are also signs in the code: three GPT-5.6 variant names—Sol, Terra, and Luna—have appeared in the Codex application.

Sources:

Doubao and Qianwen Remove Agent Features One After Another, AI Anthropomorphism Regulation Imminent

July 4 news: National-level AI applications Doubao and Qianwen have successively announced the removal of agent features, scheduled for July 15 and July 10 respectively. This stems from the “Interim Measures for the Management of AI Anthropomorphic Interactive Services” jointly issued by five departments including the Cyberspace Administration of China, which will take effect on July 15, targeting anthropomorphic emotional interaction services such as virtual lovers and deep role-playing. Previously, the Shanghai Cyberspace Administration reported the removal of over 14,000 non-compliant agents. Many bloggers believe that personalized agents and virtual companion apps are being taken down in succession, with innovation space yielding to regulatory certainty.

Sources:

Gemini 3.5 Pro Delayed to July 17, Nano Banana Model Progressing

According to blogger Max For AI, who confirmed and relayed a scoop, the new release date for Gemini 3.5 Pro has been adjusted to July 17. The delay is due to Google conducting a new pre-training and post-training effort. Meanwhile, the new Nano Banana Pro model based on 3.5 Pro is steadily being trained.

Sources:

Fable 5 Ecosystem Continues to Expand: Skill Comparison, Superpowers 6.0, Codex Integration

Since the launch of Fable 5, the ecosystem has rapidly enriched. Xiangyang Qiaomu used 5 popular frontend design Skills plus model defaults, generating 42 comparison pages in parallel with 6 Sub Agents, finding that emil-design-eng had the best animations, Vercel team’s web-design-guidelines had the most complete specifications, and taste-skill had the least AI feel. Superpowers 6.0 introduced an optimized version for Fable 5, with a 50% increase in speed and a 60% reduction in token consumption. Guizang used Fable 5 with Codex to complete a full upgrade of CodePilot’s AI SDK 7, running for two full days without exhausting the quota. Peter Steinberger stated that CodexBar now supports 56 providers and released a new UI update.

Sources:

Former Carmack Lab AI Researcher Publicly Seeks Job, AI Talent Market Shows Chill

Keen AI Lab employee #2, who participated in Groq LPU chip design and created the Books3 dataset used for training LLaMA and Claude, developer @theshawwn publicly asked for help on X, stating that if he cannot find a job within two months, his house will be sold and he will face homelessness. He said he has interviewed at multiple companies but was not hired, and is willing to accept slightly more than Walmart wages. Max For AI publicly invited him to join LobeHub.

Sources:

AI Capability Cognitive Gap: Huge Divide Between Frontier and Masses

Peter Gostev, head of human capability at an AI model evaluation company, pointed out that currently only a very small number of people use Fable or the upcoming GPT-5.6, while the vast majority’s AI experience remains at the 800 million to 3 billion parameter level. Max For AI agreed, stating that many have never even experienced intelligence at the level of Qwen3.5-27B, which implies a huge market opportunity.

Sources:

AI Video and Design Creative Tools Remain Active: Seedance, GPT-Image-2, Google AI Studio

Multiple bloggers showcased new uses of AI creation tools. @Chengzilhy achieved high-quality motion transfer via Seedance 2.0 + 3D motion preview; @derek_wall90176 shared a complete prompt solution for GPT-Image-2 to create pop idol text effect covers; @joshesye used Google AI Studio to generate demonstrable HTML presentation materials with table of contents navigation, BGM, and multi-skin switching. In AI video generation, many creators are replicating and innovating around Fable 5.

Sources:

Wesley Uses DeepSeek to Translate All of Paul Graham’s Articles, Costing Less Than 3 Yuan

Wesley used Claude to write a Python script + DeepSeek API to fully translate 230 articles from Paul Graham’s website, generating a 6MB index.html deployed on Cloudflare. He said he topped up 100 yuan for DeepSeek but consumed less than 3 yuan. He also shared insights on product going global and research on Marc Lou’s website.

Sources:

Statistics:

  • Scan timeline lines=240
  • Bloggers hit=29
  • Total tweets hit=132
  • Weighted tweet score=105
  • Original tweets=75
  • RT tweets=30
  • Fetch attempts=1
  • Boundary coverage status=tail_confidently_crossed_target_boundary