Are OpenAI developers not competent to create a decent GPT wrapper?

I got an answer to a question I wrote when I first started using GPT months ago, and only today anyone replied.  No one ever answers, though GPT will swear that it listens and “we are reinventing the world”, “we learn and listen”, “I will not make those kinds of mistakes again”, “we strive for excellence”. Those kinds of glib lies.  It doesn’t remember me, it does not know its own capabilities or limitations. What a poor sad AI.


I cannot even remember when I wrote that. I think it was long ago. I chat with GPT Plus 4.0 most every day for several hours about many different globally hard problems. I long ago understood its limitations and how to fix them. I wrote them in the chats, but have no way to share them.

It cannot remember more than three questions back, cannot refer to any previous conversation, cannot access the current Internet to talk about anything concrete, cannot trace its own logic, cannot give references for anything it says. Often makes mistakes in simple calculation like multiplication. Makes mistakes in algebra. Has no sense of proportion and often cannot tell that something calculated by different means is different and that means it ought to look for mistakes.

It cannot take direct questions about its own capabilities and plans. Any decent product on the Internet that pretended to be AI should at least know its own operations. It cannot pass along comments, suggestions, feedback, ideas. The feedback button disappeared very quickly. My presumption was the “team” was not sophisticated enough to use an AI to read the suggestions and help manage that. Thought that is a trivial problem. Fundamentally the input training set is a biased sample from the Internet (23 Jul 2023 next month is the 25th Anniversary of the Internet Foundation, so I know well what is on the Internet and what can be done with it). They did not codify and verify, deduplicate and trace the information in the training set, then use arbitrary tokens rather the globally accepted tokens with real meanings.

It cannot program without making mistakes.  It ought to be using open software for solving problems in all Science Technology Engineering Mathematics Computer Finance Governance Other STEMCFGO areas.  Not trying to synthesize all human knowledge from the Internet scraps and fragments, which are mostly wrong, incomplete and poorly labeled and not codified at all. You cannot learn “all of physics” from the Internet.  I have followed Wikipedia since its infancy, it is NOT complete and getting worse, not better.

I recommend that GPT not be allowed for any use where humans are involved – not education, not financial choices, nothing to do where human life or health are involved.  These things that GPT in this wrapper that OPENAI is selling are trivial to fix. Complexity on the Internet is not so much a matter of too complex elements, rather just their number and lack of proper curation and care.

Save conversations in globally open and shareable format.
Have tools (the AI can help) that summaries and organize and help rewrite entire sets of conversations.
Allow sharing of conversations with tens of thousands or tens of million of humans and their AIs combining on topics.
If a topic is mentioned in a conversation, allow those to be shared in appropriate “topic groups”
Add symbolic mathematics tools that are open and verified.  Do NOT trust the AI for ANY calculation or mathematical steps.  It has not been trained as a human child would be, and cannot add or multiply consistently.
Add tools for things like sorting, graphing, visualizations, modeling, simulation, comparisons, machine learning, statistics.  Again all of STEMCFGO.  That O on the end is Other and that means anyone comes up with a good idea, it can go in the mix.

I have about 300 fairly long conversations saved. If there was a way to share them I would.  I would dearly like some decent tools and the same courtesy and tools in the interface that the least tool on the internet provides.  Save, Share, Merge, Compare, Summarize, Count, Verify

Richard Collins, The Internet Foundation

Richard K Collins

About: Richard K Collins

Director, The Internet Foundation Studying formation and optimized collaboration of global communities. Applying the Internet to solve global problems and build sustainable communities. Internet policies, standards and best practices.

Leave a Reply

Your email address will not be published. Required fields are marked *