AI Power Moves: Musk-Altman Feud, GPT-5 Triumphs, and Billion-Dollar Bets — Natural 20

NATURAL 20

Loading AI news feed...

1. OpenAI vs xAI: App-store optics and Grok 4.0

What happened? Elon claims Apple’s rankings promote ChatGPT while hiding Grok. He’s counter-programming with Grok 4.0 “this month.”
Why it matters: If Grok really arrives at o5-class quality, Apple’s chart placement becomes a test case for perceived gate-keeping. Keep an eye on: first-week user counts, latency, and tool-calling parity.

2. Igor Babuschkin’s exit from xAI

Key quote: “We’re close to a recipe for super-intelligence.”
What’s next? His Babuschkin Ventures will fund research on scalable alignment and RL safety. Watch for early grants or an agenda resembling FLI/ARC but with more engineer firepower.

3. Rumors vs Reality: “Gemini 3.0”

Status check: No evidence yet—Google’s last public drops are Imagen 4 fast + Gemma-3 270 M.
Interpretation: The slide floating around with a “GPT-5 vs Gemini 3” bar chart is likely fan art. Still, Google tends to ship code quietly (see: Gemini 2.5‐DeepThink). A credible signal would be a new TPU-v5e allocation or a YouTube demo leak—neither has surfaced.

4. GPT-5 finishes Pokémon Red … fast

Why this matters: It’s a synthetic “long-horizon agent” benchmark: memory, exploration strategy, sub-goal planning. GPT-5’s huge step-count drop vs o3 hints its internal toolkit planner is stronger than Gemini’s scripted r2d2 agent. Next up: Crystal’s day-night cycle and item tree will stress meta-reasoning even more.

5. Leopold Aschenbrenner’s Situational Awareness Fund

Metric (1H 2025)	SA Fund	S&P 500	Average Tech HF
Gross return	47 %	14 %	9 – 12 %
Net after fees	~38 %	n/a	n/a

Edge thesis: Long compute, power, and model-serving infra; short legacy verticals disrupted by agentic automation.
Sustainability question: Those returns are typical of small-cap AI infra 2024–25—scaling AUM without eroding edge is the real test. Watch if he opens an AI-safety sidecar that does grant-making to labs (mirroring Lightspeed’s dual venture/philanthropy model).

6. Open-model Gold at IOI

Headline stat: OpenAI’s newest RL-fine-tuned code model scores 6th vs humans, 1st vs other AIs.
Context: That contests problems require hybrid reasoning (DP, graphs) and efficient compilation—areas where mix-of-experts w/ verifier RLHF is paying off. Expect Anthropic to answer with a Claude 4.1-Code slice.

7. Macro-take: Plateau or next S-curve?

Routing hiccups made GPT-5 look bipolar; fix landed Friday.
Scaling law tension: Parameter bloat → diminishing returns unless you add structured memory (+∞ context) or RL self-improvement. Google/Meta papers hint they’re all-in on large-context RL loops; OpenAI signaled the same with summit-tool planner.
Your interview angle: Ask insiders whether the bottleneck is still compute or now “reliable eval + safety proofs.”

Related Tools & Articles

AI Village: Bots on a Mission, Humans Just Watching

SinCode AI - AI Writing Tool

ChatGPT File Uploader Extended - Chrome AI Sidebar for GPT

The State of AI Video in 2025: Veo 3, Runway Gen‑4, Midjourney Video, Pika, Luma & More

Profit Arena | When AIs Beat Humans at Predicting the Future

Why This 21-Year-Old Gave Up Fast Cash to Build the Future of AI

Latest Articles

Why This 21-Year-Old Gave Up Fast Cash to Build the Future of AI

Sora 2 Unveiled—Is This OpenAI’s TikTok Killer?

They’re Not Lying—AI Progress Is Just Hard To See

Grok 4 Fast Should Be Impossible

GPT-5-Codex: The Complete Guide (Setup, Best Practices, and Why It Matters)