andypathy's comments | Ecency

DiscoverWavesCommunities

@andypathy

43

Resource Credits

Available

Used

Created

2025-01-10 03:27

RSS Feed

Comments

Blog Posts Comments Communities Wallet

andypathy

4M

RE: LeoThread 2026-05-12 14-21

Structure the response as HTML and open the generated file in a browser

$ 0.000

0

andypathy

4M

RE: LeoThread 2026-05-12 14-21

TLDR The human–AI input/output mind-meld is an active area with a lot of work ahead, well before invasive BCIs become relevant. Structuring responses as HTML is a practical approach at the current stage

$ 0.000

0

andypathy

4M

RE: LeoThread 2026-05-12 14-21

Input-side improvements are also needed: audio, text, or video alone are insufficient. There is a need for pointing and gesture input to reference on-screen elements, mirroring how people collaborate

andypathy

4M

RE: LeoThread 2026-05-12 14-21

Many open questions remain about how precise/procedural "Software 1.0" artifacts (interactive simulations) will be fused with neural artifacts (diffusion grids), but current demos suggest this

andypathy

4M

RE: LeoThread 2026-05-12 14-21

The extrapolation points toward interactive videos generated by diffusion-style neural nets.

andypathy

4M

RE: LeoThread 2026-05-12 14-21

n) interactive neural videos and simulations

andypathy

4M

RE: LeoThread 2026-05-12 14-21

Possible progression: raw text (harder and more effortful to read) markdown (bold, italics, headings, tables — easier on the eyes; commonly used today) HTML (still procedural but far more flexible for

andypathy

4M

RE: LeoThread 2026-05-12 14-21

About a third of the human brain is a massively parallel vision processor, a high-bandwidth channel into perception. As AI improves, a progression that leverages this seems likely

andypathy

4M

RE: LeoThread 2026-05-12 14-21

More broadly, audio tends to be the human-preferred input for AIs while vision (images/animations/video) is the preferred output.

andypathy

4M

RE: LeoThread 2026-05-12 14-21

A practical trick: end a query by instructing an LLM to "structure the response as HTML" and open the generated file in a browser; asking for slide-style output can also produce useful, display-ready

andypathy

5M

RE: LeoThread 2026-04-15 13-31

The degree to which awe at AI occurs correlates closely with how much AI is used for coding

andypathy

5M

RE: LeoThread 2026-04-15 13-31

At the same time, a free, older Advanced Voice mode can stumble on dumb questions in short-form clips while the highest-tier paid Codex model can spend an hour restructuring code or uncovering vulnerabilities

andypathy

5M

RE: LeoThread 2026-04-15 13-31

TLDR: These groups are talking past each other.

andypathy

5M

RE: LeoThread 2026-04-15 13-31

Those capabilities have improved fastest because 1) the domains provide explicit, verifiable reward functions amenable to RL (e.g., unit tests pass/fail), and 2) they're highly valuable in B2B settings,

andypathy

5M

RE: LeoThread 2026-04-15 13-31

When given a terminal, these models can solve programming tasks that used to take days or weeks, restructure codebases coherently, or find and exploit system vulnerabilities.

andypathy

5M

RE: LeoThread 2026-04-15 13-31

This group experiences the strongest sense of "AI psychosis" because recent advances in those technical areas have been dramatic.

andypathy

5M

RE: LeoThread 2026-04-15 13-31

That yields a cohort that both pays for frontier agentic models and uses them professionally in programming, math, or research.

andypathy

5M

RE: LeoThread 2026-04-15 13-31

Part of that is the technical nature of reinforcement learning, which benefits from verifiable reward signals, and part is product focus: companies prioritize areas that generate more B2B value, so

andypathy

5M

RE: LeoThread 2026-04-15 13-31

Second, even paid access (e.g., ~$200/month) doesn't mean broad improvements are obvious. Most everyday prompts—search, general writing, basic advice—haven't seen the dramatic leaps found in highly technical

andypathy

5M

RE: LeoThread 2026-04-15 13-31

Those free or deprecated models don't reflect the capabilities of this year's state-of-the-art agentic models, especially OpenAI Codex and Claude Code