{"id":14733,"date":"2024-04-22T15:47:11","date_gmt":"2024-04-22T15:47:11","guid":{"rendered":"\/?p=14733"},"modified":"2024-04-22T15:55:08","modified_gmt":"2024-04-22T15:55:08","slug":"llm-wrapper-programs-llm-shells-limited-closed-staffs-shallow-internet-skims-need-proper-tools-for-true-ais","status":"publish","type":"post","link":"\/?p=14733","title":{"rendered":"LLM wrapper programs, LLM shells, limited closed staffs, shallow Internet skims -> need proper tools for true AIs"},"content":{"rendered":"
\n
\n
@AIatMeta<\/span> I talked to AI at Meta today. I spent thousands of hours interviewing these “LLM wrapper programs” and all of them behave almost identically. I was interviewing them to see if they would be capable of helping with any of thousands of global and systemic issues I have identified in the last 26 years of the Internet Foundation.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
Meta AI (a wrapper program for a closed LLM based on unindexed input data with no ability to trace it sources):<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
It has no permanent memory so it will not remember conversations.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
It is not allowed to access the Internet so it will not be able to locate resources and complete answers to anything.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
Is based on a shallow skim of data from the Internet, biased to only “free” materials, ignoring all the copyrighted material, much which forms the basis of our global human society.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
It will not use a calculator when it does calculations. It will not use symbolic math tools which are known to be correct – rather it takes the shallow skim of the Internet and tries to learn, on the fly, how to solve mathematics problems that have taken humans centuries and much creativity.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
It cannot (is not allowed) to us computer software and systems to investigate and research on behalf of human clients. So it will only tell you guesses from the LLM statistical index how to program, and will not write and test the programs for you.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
The Meta AI, OpenAI ChatGPT, Google Bard\/Gemini, Microsoft Bing CoPilot, and other self-proclaimed “AIs” are only wrapper programs for LLMs. I calll them “LLM shells” — written by small groups inside for-profit corporations. With insufficient experience, closed to outside suggestions.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
Meta AI will NOT take specific feedback on bugs, nor suggestions, nor feedback. It is a shell, not a real thing.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
I can be more specific. The world runs now mostly on lossless indexed databases, and is not using the LLM statistical indexes at all, except for shallow games and demos. It is possible to find a place for LLMs, but it requires a much higher quality of openness, trust, verification, public and government audit and oversight, global collaboration. I have written most of that down. <\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
Conversation should be stored losslessly and permanently. The humans should be able to mark what can be share globally, in groups, or not at all. If “not at all” that means “no casual and un-tracked browsing by Meta (or other company) staff or contractors.”<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
I am working long hours 7 days a week to check what can be done. So this is just a short note to mention the most egregious failings of Meta. Meta copied OpenAI almost exactly and kept every one of the failings of its “LLM shell program” and related policies and bad Internet business practices.<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
I am fairly certain that the proper path will be to complete the words of Meta and others about “all human languages” and do it right – for the whole Internet, not just another public relations sales demo. Then tackle “all knowledge” but where the LLM is just one of the tools used by “AI shell programs”, including where the shells call many LLMs optimized for different real tasks..<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
Filed under “LLM wrapper programs, LLM shells, limited closed staffs, shallow Internet skims -> need proper tools for true AIs”<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
Richard Collins, The Internet Foundation<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
\u00a0<\/span><\/div>\n<\/div>\n<\/div>\n
\n
\n
[ A bit harsh but true: The “GPTs” of OpenAI are a sham (and a shame). The groups offering add-ins are at the mercy of the limited shell program and the narrow policies and limitations of a few individuals. I have a specific plan now, based on a lifetime (60 years) of solving these kinds of systemic problems. ]<\/span><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

@AIatMeta I talked to AI at Meta today. I spent thousands of hours interviewing these “LLM wrapper programs” and all of them behave almost identically. I was interviewing them to see if they would be capable of helping with any of thousands of global and systemic issues I have identified in the last 26 years
Read More »<\/a><\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[73,72],"tags":[],"class_list":["post-14733","post","type-post","status-publish","format-standard","hentry","category-all-knowledge","category-all-languages"],"_links":{"self":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts\/14733","targetHints":{"allow":["GET"]}}],"collection":[{"href":"\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=14733"}],"version-history":[{"count":4,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/14733\/revisions"}],"predecessor-version":[{"id":14737,"href":"\/index.php?rest_route=\/wp\/v2\/posts\/14733\/revisions\/14737"}],"wp:attachment":[{"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=14733"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=14733"},{"taxonomy":"post_tag","embeddable":true,"href":"\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=14733"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}