“Its the great AI Turing-Test-Off!
I asked Claude, Bing, Bard & GPT-4 to ask a question, and guess which answer was written by me, and which was written by one of the other AI models
Bing & Claude both thought GPT-4 was human. GPT-4 thought Bard was. No one thought I was human.”
I asked Claude, Bing, Bard & GPT-4 to ask a question, and guess which answer was written by me, and which was written by one of the other AI models
Bing & Claude both thought GPT-4 was human. GPT-4 thought Bard was. No one thought I was human.”
❤1😁1
Twitter Releases (Partial) Recommendation Algorithm Code
Diagram shows the usual high-level pipeline:
• Raw data,
• Unsupervised feature-engineering,
• Quick rough candidate pre-selection,
• Slower in-depth reranking,
• Mixing in other types of recommendations.
Ok, normal so far. Gonna have to dive in to find any more interesting parts.
Announcement
Github
Diagram shows the usual high-level pipeline:
• Raw data,
• Unsupervised feature-engineering,
• Quick rough candidate pre-selection,
• Slower in-depth reranking,
• Mixing in other types of recommendations.
Ok, normal so far. Gonna have to dive in to find any more interesting parts.
Announcement
Github
👍6🤯3❤1🤬1👀1
Why doesn't twitter just release their algo's high-level equation?
Fun fact: Our current math lacks the language to even do this.
I.e. the usually-used probability pseudo-notation neither can express neither the questions nor their answers, for many of most common basic high-level, practical questions one tends to have about modern AI models and their training. Several basic notions are totally absent, inexpressible in the current math notation.
Not just a problem of missing answers, but inability to even concretely express the questions.
Even worse, the notation that is used isn’t just incomplete but outright wrong. Technically, machine learning systems badly-violate many of the basic conditions neccesary for the usually-used probability notation the be valid. Never was really even really in question, the probabilty-based notation was always a lie.
Why do experts joke that we don’t really understand deep learning AI all? Missing math language.
I.e. idk lol.
Fun fact: Our current math lacks the language to even do this.
I.e. the usually-used probability pseudo-notation neither can express neither the questions nor their answers, for many of most common basic high-level, practical questions one tends to have about modern AI models and their training. Several basic notions are totally absent, inexpressible in the current math notation.
Not just a problem of missing answers, but inability to even concretely express the questions.
Even worse, the notation that is used isn’t just incomplete but outright wrong. Technically, machine learning systems badly-violate many of the basic conditions neccesary for the usually-used probability notation the be valid. Never was really even really in question, the probabilty-based notation was always a lie.
Why do experts joke that we don’t really understand deep learning AI all? Missing math language.
I.e. idk lol.
👍7🤣5❤1