Free — no credit card required
Know what to watch.
In 5 minutes a day.
Alize watches 100+ YouTube channels across AI, dev tools, startups, and more — then emails you only the videos that matter. Every morning.
Free forever·No credit card·Set up in 30 seconds
100+
Channels monitored
Daily
Morning delivery
5 min
Time to stay informed
Free
No credit card
This is what you get
Not summaries. An actionable briefing.
Key takeaways, why it matters, and what to do — for every video worth your time.
alize.me · Morning digest
Alize
Your daily digest
Today · 8
Archive
Channels
Settings
8 videos worth your time
AI/LLMs, Startups, Dev Tooling
AI / LLMs
Andrej Karpathy
Tokenization is a deeply strange problem — here's why it matters
Karpathy argues that nearly every LLM failure mode — from bad math to broken code — traces back to tokenization artifacts. Proposes a new byte-level approach that could eliminate these issues entirely.
Key takeaways
- GPT-4 fails at simple arithmetic because numbers tokenize inconsistently
- Byte-level tokenization eliminates 90% of encoding edge cases
- Google's Gemini already uses a hybrid approach internally
Why it matters
If you're building on LLMs, tokenization bugs are silently degrading your outputs. Understanding this helps you write better prompts and debug faster.
What to do
Test your critical prompts with the OpenAI tokenizer tool. If key terms split across tokens, restructure your input.
Startups & Business
Y Combinator
Why the new YC batch is betting on AI infrastructure, not apps
The W26 batch reveals a dramatic shift: 60% of funded companies are building infrastructure, not consumer apps. Garry Tan explains why the "picks and shovels" thesis is winning.
Key takeaways
- 60% of W26 batch is infra — up from 35% in W24
- AI app churn is 70% within 3 months vs 30% for infra tools
- Three funded companies are building open-source alternatives to closed APIs
Why it matters
The smart money is moving downstream. If you're building an AI app, your competitive moat just got thinner — infra is where defensibility lives.
What to do
Evaluate whether your product relies on a commodity AI layer. If yes, consider what proprietary data or workflow lock-in you can build.
How it works
Three steps. Five minutes a day.
01
Pick your channels
Choose from starter packs or add your own YouTube channels.
02
AI reads everything
We fetch transcripts and score each video on relevance, novelty, and signal-to-noise.
03
Get your morning briefing
Only the videos worth your time, with key takeaways, why it matters, and what to do.
Starter packs
Pick a pack, start in seconds
Curated channel bundles for every interest.
Stop scrolling. Start knowing.
Your morning briefing is waiting. Free, no credit card, set up in 30 seconds.
Get your first digest