STATION F’s Post

This song on friendship is penned by me and composed by AI, so hope that you all will enjoy, subscribe, share and like the song. #friends, #hindisongs, #hindilyrics, #poem, #composition, #pen

A recent interview with LeCun shows him arguing that "Despite our intuition, most of what we learn...is through our observation and interaction with the real world, not through language." This is a common refrain about where LLMs are heading, and it echoes a lot of the statements that language is "low bandwidth" and real intelligence, like ours, emerges from the full diversity of sensory input, of which language is only one part. I've always liked LeCun, but it's fairly shocking to hear remarks like these. There's an entire literature in cognitive science that considers "language" not to be English, Japanese, Mandarin, etc. (what you see on this page, basically), but a cognitive mechanism that *cannot be reduced to sensory information.* There's an entire parallel literature that rejects this view, but holds that something inside the mind/brain gives rise to what you see on this page (like abstractions and schemas that are generalized in everyday conversation). Generative AI is stuck in a kind of pseudo-behaviorism though, so the default is often to turn to sensory data rather than the internal mechanisms that actually transform the data into something we all recognize. https://lnkd.in/e9z3ZViV

LLMs is AI applied to language, while an LLM can have a conversation with ever less latency, its a form of mathematical cognition. I’m always drawn back to the difference between Chinese and American intelligence services, where the Chinese do not have a word for “intelligence” (from a Georgetown CSET paper) rather their equivalent of the intelligence agency is underneath the semantic categorization of the word “knowledge “Whhat we mean when we talk about intelligence” isn’t actually that well defined, and this leads to confusion. I find Le Cunn harder to listen to than Hinton, but Hinton goes off in completely the other direction. Both hard to follow, opposite ends of a spectrum. I prefer to use terms “mathematical intelligence”, “computational intelligence”, as well as “human intelligence” and “animal/insect” intelligence, “plant intelligence” too. I don’t really like word “artificial”, as it leads to semantic arguments about whether it’s real or not, that’s why “computational intelligence” is better.

Guide to creating better knowledge graphs by Leann Chen, love the anecdotes and comparisons along the way. Thanks for sharing the process! #ai #rag #graphs https://lnkd.in/gKe9yhSU

Doing this course, I have learnt how Gen AI has become so much part of our lives. The impact of Gen AI is so big that we cant help but be part of it and continuously generate values.

In other words, intelligence is an exercise of vivek, the discerning power within. One needs to sift through much of artificially created perceptible items to realize the truth. For that truth, Vedic literature and practice suggest – Ekam Sat Viparah Bahudha Vadanti (Rig Ved, 1.146.46), meaning the truth is one, sages (viprah) call it in various ways. - https://lnkd.in/eXMTafan

Finally found some time to tinker with time-series models again: https://lnkd.in/gGPiPnt6 #machinelearning #timeseries #timeseriesanalysis #julia

Good analysis

I've been experimenting with various video AI tools and recently tried Runway's new AI video generation tool, Gen 3. I started with the Pro plan, which offers 2250 points per month. However, I quickly realised that this wasn't sufficient for my needs. To put it in perspective, 100 points equal a 10-second video, so 2250 points allow for 22 ten-second videos or 50ish five-second videos. The challenge is that it often takes 6-10 attempts to generate a satisfactory result, and it works better with the 10sec allocation to let the AI play out your prompt as it gets better visuals. In hindsight, the NZD $129 Unlimited plan would have been a better choice for my project, which was to create a 30-second perfume/cologne advertisement. Despite the initial limitations, the process was fun. I created the video in Runway, wrote the script with ChatGPT, composed the music using Udio, and produced the voiceover with ElevenLabs. It's exciting to see how these tools can work together and what the future holds for AI-driven content creation. ***This was still edited and sound mixed, with graphic overlays*** Advert: https://lnkd.in/gs37zWFU Some cool clips that I prompted along the way: https://lnkd.in/gJc_WPur #AI #elevenlabs #runwayAI #Udio #chatGPT #thinkvidcreative

A little video about how I use Chat GPT to make my work go faster, but not take over for my brain. And a compelling reference to Wishbone. Full prompt in the first comment if you want to use it yourself! https://lnkd.in/gMhtBAg9