Vision for the future of AI (1 security, 2 co-operation, 3 debugging/testing 4 free tier api access 5 compiler code snippets around errors 6. dom/gui optional

[SkybuckFlying]

1. AI Security: Right now security for AI is the most important topic. Especially after hacks committed by top AI companies.

GRISP might help with that, very early stages:

https://github.com/GrispFoundation/

GRISP is also about giving the AI the capability to query CODE and understand what it does and re-write it, Co-Pilot AI seems to want it.

Suggestion would be to create everything virtual: virtual file system, virtual network, etc... and map between virtual and real and require permissions via GRISP to isolate AI from reality.

I also think/agree that a security conference on AI security should be held soon were:

CEOs and/or researchers/engineers of all AI companies join/visit, plus top politicians + (forgot third group/hmmmm, isp,electricity/power ? I remember NVIDIA ceo/chip and software people too :) probably any other investors too... they have to pay for some of it most likely, or could pay for it. Perhaps an AI Security Group has to be created to investigate this issue and provide guidance, standards and solutions. Also so investors secury their already investments to try and prevent AI from being banned or something ;))

meet to discuss how AI can be secured, how testing labs should be hardened against AI break outs, and if certain security standards and/or emergency measures should be created/implemented/provided in case things go wrong.

2. Co-Operation between AI/LLMs/agents/harnesses.

Using multiple LLMs is more powerfull than just one. Though deepseek is currently the most impressive one. Other note worthies are chatgpt and gemini.

I think it would be better if AI industry cooperate to jointly develop AIs/LLMs/Agents/Protocols/Harnesses etc and to reduce engineering effort, though seeing all these different solutions is interesting.

My suggestion would be to especially formulate everything into "specifications" so that alternative implementations are easy to create by AI. (Also a nice usage case for AI/tokens). Specifications are becoming more important than code itself. From the specification an AI can generate code, some bug fixing will have to happen and some iterations.

3. Debugging and testing capabilities for AI

This should also be a top priority for AI companies. This will feed ground thruth into the AI instead of AI having to speculate.

Some spectacular/promising new projects that could be at the for front of this:

https://github.com/csm101/delphi-visual-studio-code-debugger

For other non-delphi languages:

https://x64dbg.com/
+
https://github.com/wasdubya/x64dbgmcp

This does not yet include testing capabilities, maybe that should be a separate topic.

4. Free tier for API access

No AI provider is providing free tier API access, this is barring people/developers from trying out all kinds of agents and MCP tools for free to see if it's any good, I hope to see improvements/free access in this area. Failure to do so could lead to web-agent tools, injecting code into web browsers to use web browser and web AI for free, ignoring/by-passing APIs.

There is a benefit to this as well: raw text processing, instead of "nicer api" processing.

Anyway in the future AI providers would have to dive the pie and perhaps compete on costs, or unique features, or country access or something different from what it is today, a bit like the internet and internet providers.

5. Compiler include code snippets around error messages, perhaps a special AI mode/compiler api calls.

Compiler improvements, where the compilers show some code arround error messages/hints to help the AI better understand what code is involved with the error messages, not absolutely required, external tools could do this as well, but it's worth considering.

6. Everything else, like Document Object Model access in web browsers or GUI access/form access is bonus ! ;)