techFri 2 Oct, 07:30

Sources93 X accounts · techtwitter.com · Hacker News · 8 newsletters
  1. Runs every day at 07:30 London time.
  2. It is shown its own papers of the last 3 days, to know what it has already said.
  3. Haiku 5.5 writes the paper from what was fetched, with no tools.
  4. Jev scores every story by how much it matters to you; each section puts those first, and what is not for you drops to the briefs.

93 X accounts

Their posts of the last 24 hours, through FxTwitter; plus everclear077/awesome-ai-x-accounts when it answers.

techtwitter.com

What tech Twitter is discussing, ranked by discussion.

Hacker News

The front page, plus stories over 80 points in the last 24 hours.

8 newsletters

Every new issue, each delivered once, looking back up to 7 days.

The tech daily written for one reader

Claude Code mods let you customize everything by prompting

Anthropic is shipping mods as a first-class feature: small JS/TS files that run in your session, watch events, rewrite behavior, or draw UI. You don't edit code—you just tell Claude what you want and it builds the mod. Share them as plugins for others to use.

Remote video interviews are back because AI fooled the detectors

Companies are requiring on-site interviews again after AI video generation got good enough that people can no longer tell deepfakes from real video calls. Some firms are even seeing North Korean hackers attempt to pass interviews with deepfakes. The arms race between detection and generation is heating up.

Web dev education is dying

A well-written essay argues that traditional web dev education—bootcamps, courses, mentorship programs—is collapsing because frontier models can now teach and debug faster than humans. The career ladder is flattening.

Big announcements

Friday, October 2, 2026 · Pergent2

OpenAI rate-limiting Sol after demand surge

GPT-6.1 Sol was so popular it overwhelmed inference capacity. OpenAI is resetting usage tomorrow at 10am PST for all paid ChatGPT accounts and apologizing for the slow start over the last two days.

Perplexity open-sources decision model pplx-decider-27b

Perplexity dropped a multimodal decision model trained from Qwen and released it on Hugging Face (Apache 2.0). Also launching a Decisions API at $0.04 per million input tokens with free output—aiming to undercut competitors even further.

Kev 1.0: open-weight decision models you can train yourself

Jared Palmer released Kev 1.0 with new Kev-27B and updated Kev-9B models, 64k-token context window, and skills to fine-tune on your own data. TypeSafe SDK compatible. Available on Hugging Face.

LlamaIndex Extract v2.5 dominates document extraction

New series of frontier agents tuned for document extraction: cost-effective, agentic, and agentic-plus tiers. Outperform Opus 5.5 and GPT-6 Sol while costing 30%-4x less. Major improvements on long lists (86→96%), records spanning pages (86→97%), and scanned forms (91→96%).

What people are fighting about

Friday, October 2, 2026 · Pergent3

PewDiePie banned twice by OpenAI for distilling Sol

PewDiePie tried using Sol's chain-of-thought outputs to train a smaller model. OpenAI banned him, he appealed and got unbanned, then ran Sol again and got banned again. The incident raises hard questions: do model outputs count as intellectual property? Can users legally distill?

Web dev corner

Friday, October 2, 2026 · Pergent4

Vector databases might be dying

Turbopuffer posted "RIP, vector database"—arguing that specialized vector DBs are unnecessary when you can do vector search directly in Postgres or any competent OLAP database. The blog sparked 300+ comments about whether dedicated systems still have a place.

From the newsletters

Friday, October 2, 2026 · Pergent5

How AI is changing cybersecurity's balance of power

Adam Meyers (CrowdStrike head of counter adversary ops) explained on The Deep View Conversations that AI is helping defenders match the speed and scale of attackers. Key insight: the harness around the model matters as much as the model itself. Changing harness parameters reduced false positives without touching the underlying model.

Research & open source

Friday, October 2, 2026 · Pergent6

Worth a read

Friday, October 2, 2026 · Pergent7