3b‐1 Agent concepts2 - terrytaylorbonn/auxdrone GitHub Wiki

26.0318 Lab notes (Gdrive) Git

This page describes my take on core AI agent concepts.

• 1 Agentic AI "on rails" is the future of AI
• 2 Input junk into a transformer/CNN and you get junk output. Intelligence?
• 3 End users need a realistic (honest) assessment of AI
• 4 AI will never have intelligence, so how to move forward? Limit ("harness") AI inputs
• 5 My demo roadmap
• 6 My Substack posts
• 7 Youtube videos (4 YT videos that give great insight into agents) (all have English subtitles and audio available)

Constraining model inputs (harness engineering, AI on rails) is the only way to avoid junk AI outputs.

The term "on rails" is borrowed from Ruby on rails.

• Input junk and AI outputs the most probable match.
• Input something thats not junk, but for some reason the UFA does not approximate correctly, then the result is junk.
• AI does not warn you when it outputs junk.

I have been focusing a lot recently on my long-standing claim that GPU-based AI has no intelligence and never will. The reasons for this

• 1 Honesty pays (especially if you are an AI end-user).
• 2 AI gurus know that AI has no intelligence.
• 3 To truly exploit AI, you need to have a realistic assessment of its capabilities.

Slowly the limitations of AI (it is not intelligent and never will be) are being acknowledged. Part of the reason is the rapidly growing customer base is accumulating a lot more hands-on experience.

• Create a “sandbox” for the LLM that will limit the input to the LLM.
• This will be analogous to binary apps running on an OS “sandbox” (preventing the app from crashing the OS).

This is still under development. I will start on this on 26.0326. Wiki page: https://github.com/terrytaylorbonn/auxdrone/wiki/3b%E2%80%902-Demo-roadmap .  

• #70 “AI is simply a new tool to allow humanity to fly even higher, even faster (and, perhaps, to fall even harder)
  • This post explains (with analogies) the core differences between AI and real (human) intelligence (v2)”
  • https://ziptieai.substack.com/p/70-ai-is-simply-a-new-tool-to-allow
• #69 "The future of AI platforms (AIPs)
  • AIPs will always be binary agentic deterministic programs wrapped around an AI NN (such as an LLM)"
  • https://ziptieai.substack.com/p/69-the-future-of-ai-platforms-aips
• #68 "What is an AI platform?
  • Basically a new iteration of the traditional app/POSIX/OS architecture"
  • https://ziptieai.substack.com/p/68-what-is-an-ai-platform

Recent videos from my favorite AI blogger “Best Partner” that sum things up perfectly.

• 7.1 A VC analysis of the future agent market (it belongs to those who adapt quickly) (#375)
• 7.2 The future of AI belongs to framework tools that maximize AI model reliability (such as Harness) (#377)
• 7.3 An AI guru talks about the “bitter lesson” (bitter for AI gurus) of the unrealized promises of AI (#Welch)
• 7.4 An AI guru who has far more modest expectations (promises) for AI (#376)  
  
  

https://www.youtube.com/watch?v=mNhHjkf_ahM 【人工智能】Cursor要凉了么 | 顶级风投杰瑞·默多克警告过时 | 自主Agent | SaaS海啸主浪潮未至 | 编排层革命 | 软件买家剧变 | 基础设施重构 | AI原生公司 | 科技趋势

【人工智能】Agent Harness Engineering | Agent驾驭/管控工程 | 长时任务的缺陷 | 计算机的操作系统 | 通用型和垂直型 | 苦涩的教训 | 工程实践 https://www.youtube.com/watch?v=qua6FfJmydo

The following is a quote from Richard Sutton (AI guru) in the recent YT video “Can humans make AI any better?” (from Welch Labs https://youtu.be/2hcsmtkSzIw?t=547 ). This article states that it was a bitter lesson to discover that the “contents” of minds are so “irredeemably” complex (intelligent) that AI can not fake human intelligence. The answer: Use the same AI tech to search and discover the world by itself.

【人工智能】AGI还早着呢 | 吴恩达 | Agentic AI | 规模化Scaling Laws瓶颈 | 持续学习难题 | 灾难性遗忘 | 开源模型优势 | 赋能人类 | 职业重构 | 行业泡沫

https://www.youtube.com/watch?v=D5rlQiSvbek