Episode Details
Back to EpisodesAI Daily for 22 September: Grok 4.7, Kev Decision Models, Mini-AGI, macOS AI Storage
Description
AI Daily for 22 September recaps 5 major AI Hacker News stories, moving through grok 4.7, kev decision models, mini-agi, macos ai storage.
Chapters
- 00:00:00 — Intro
- 00:00:12 — Grok 4.7
- 00:01:22 — Kev Decision Models
- 00:02:24 — Mini-AGI
- 00:03:46 — macOS AI Storage
- 00:04:59 — AI-Powered CI
- 00:06:17 — Closing
1. Grok 4.7
The next story is Grok 4.7, which xAI presents as its most capable model for coding and knowledge work, trained with a longer reinforcement-learning run, better self-checking, and frontier price-performance at $2 per million input tokens and $6 per million output tokens. The Hacker News reaction praises its research and frontend work but criticizes early task termination and coding reliability, while questioning how well benchmarks reflect real-world use.
2. Kev Decision Models
The next story is Kev, a family of small decision models built on Qwen3.5 that can answer yes-or-no, multiple-choice, and rating questions with probabilities, giving software a flexible way to make fast decisions. The Hacker News debate focused on routing, triage, games, and coding-agent control loops, alongside skepticism that a new model family would beat established classifiers or strong language models.
3. Mini-AGI
The next story is a Show HN project called Mini-AGI, a byte-level language model that trains continually from a stream of data on a single 8 GB VRAM GPU, with its author claiming that the architecture can grow and prune its pool of experts while reducing catastrophic forgetting, a design that could let people train and keep improving models on their own hardware. Hacker News comments mixed excitement about locally owned AI with skepticism that this small, undertrained model had shown meaningful generalization or solved catastrophic forgetting.
4. macOS AI Storage
The next story covers a macOS 27 workaround that claims users can avoid downloading Apple’s local AI models and save up to 14 gigabytes, which matters on Macs with limited storage. Hacker News reaction centered on forced AI, uncertain Siri demand, privacy, and whether the new features justify taking away user control.
5. AI-Powered CI
The next story is Linear’s report on reworking its continuous integration system after coding agents made validation the new bottleneck, claiming it cut pull-request wait time from more than six minutes to just over five and roughly halved runner time per test. Hacker News debated how broadly those gains apply, focusing on Bazel’s setup costs, warm-cache assumptions, network and disk I/O, and how much quality assurance can be automated.
That's it for today.