AI & Tech Developments - Mar 02

GPT-5.4 looks imminent. Rumors point to a ~2 million token context window, real persistent state across sessions, and a redesigned memory architecture built for autonomous agents.

— @VraserX

The LLM form factor I think ultimately cannot escape this problem: compaction runs 5 commands 250k tokens used shit gotta compact again.

— @mike64_t

The most terrifying part of this entire sentence are the words “naturalized US citizen”. We let this guy into our country, handed him paperwork that said “congratulations, you’re an American now!”

— @ChristianHeiens

RIP Ayatollah Khamenei you would’ve loved Claude Opus 4.6 thinking mode.

— @ParikPatelCFA

Larry Ellison on the AI moat: AI is commoditizing because models use the same public internet data. The true competitive edge isn't the model itself anymore, but access to exclusive, proprietary datasets.

— @rohanpaul_ai

This GitHub repo just changed how I use Claude Code forever. It's called claude-code-best-practice and I'm annoyed I didn't find it sooner.

— @ihtesham2005

The paper says the best way to manage AI context is to treat everything like a file system.

— @rohanpaul_ai

This is completely retarded. Every "AI skill" people have learned over the last 3 years has been obsoleted 3 months later when a new model comes out.

— @hyperonline

Anthropic introduces a memory feature that lets users transfer their context and preferences from other AI tools into Claude by copying a generated prompt and pasting the result into Claude’s memory settings.

— @kimmonismus

NVIDIA just announced EgoScale. NVIDIA Research has uncovered a log-linear scaling law for robot dexterity by pretraining VLA models on over 20,000 hours of egocentric human video.

— @XRoboHub

Rumors suggest GPT-5.4 will feature a 2 million token context window and a redesigned memory architecture.

— @VraserX

Anthropic introduces a memory feature for Claude, allowing context transfer from other AI tools.

— @kimmonismus

NVIDIA announces EgoScale, a new model for robot dexterity training.

— @XRoboHub

The paper proposes a file system approach to managing AI context for better coherence.

— @rohanpaul_ai

Hyperonline critiques the rapid obsolescence of AI skills learned over the past three years.

— @hyperonline

A new GitHub repo simplifies context management for Claude Code users.

— @ihtesham2005

Mike64_t discusses the challenges of LLM context management and the need for compacting.

— @mike64_t

GenAI_is_real discusses the importance of systems knowledge in AI.

— @GenAI_is_real

arb8020 investigates the effectiveness of injecting encouraging user messages into long-horizon coding tasks.

— @arb8020

tier10k highlights the reliance on their services by sharp desks in trading.

— @tier10k

ahmetb reports on a vulnerability scanning tool's GitHub repo being compromised.

— @ahmetb

0xSero shares insights on using Qwen-3.5-Plus for coding projects.

— @0xSero