
Karpathy: Ask for Discardable Software Artifacts Now That Code Is Abundant
Karpathy says intelligence and code are abundant, so ask for custom, discardable apps and video explainers. What it means and how to try it today.

Expert profile
Founder & AI Product Leader
Yash Thakker is a Generative AI expert with over 12 years of experience in product leadership and technical strategy. As the founder of explainx.ai, he has taught over 300,000 learners and built AI platforms serving millions of users globally. He specializes in Agentic AI, Multimodal RAG, and the intersection of LLMs with consumer hardware.

Karpathy says intelligence and code are abundant, so ask for custom, discardable apps and video explainers. What it means and how to try it today.

Karpathy's Oct 2, 2026 thread ranks four LLM output formats: STE100 text, diagrams, HTML, video. Copy-paste prompts and the trade-offs.

MAI-Transcribe-2-Streaming leads AA-WER at 2.5% WER / 0.13s. Streaming costs $9/hr — more than ElevenLabs, not 60% cheaper.

Dots is always-on Astra that bills Codex/Work. Pro 200 halves on Oct 30. Tibo reset paid ChatGPT Oct 3 after 6.1 Sol load. Read the three together.

How Claude Opus 5.5, agent fleets, and VOC embeddings surfaced a 1615 dodo hunt — and why frontier research agents still fail at historical taste (2026).

Perplexity open-sourced pplx-decider-v1-27b and priced Decisions API at $0.04/M input, free output. Overall 85.71% vs Jev 84.51% on 11 benches.

PewDiePie says OpenAI banned him twice while he trained Ajax on Sol's reasoning. What's confirmed, what isn't, and the distillation rules.

Tavus Griffin is a video-to-video HIM. After 1 minute, 26 of 54 thought the partner was human. VideoFDB: 3.83 gen, 3.73 perc. Testers only.

Decision models score schema choices in one pass with calibrated noul, choice, and score fields. When to use Jev, Clef, or LLM logprobs.

ASI is intellect that beats the best humans across nearly every domain. Here is the definition, how it differs from AGI, and why 2026 models still fail it.

The video Turing test asks if people can tell an AI from a human on a live video call. How it works, how Tavus scored 48%, and why it's not proof.

xAI's dot.com redirect to Grok fueled a launch-day prank narrative after OpenAI Dots. WHOIS timing, dots.com, and why correcting the record is the ad.