Headworx

" Comprehension does appear to be separate from thinking " has been a revelation to me - as explained by Edward Gibson (a psycholinguistics professor at MIT and the head of the MIT Language Lab). Technically there is a separate neural network in a human brain which does language processing. And it has nothing to do with thinking. That makes a lot of sense and explains why the net output from all the AI progress has largely been negative .  AI Large Language Models (LLMs) work by finding statistical patterns in text. Most importantly - without connecting words to meaning. This becomes especially visible when LLMs often fabricate facts and figures, misunderstand questions, and exhibit biases found in its training data. But because LLMs have mastered language and thus sound very eloquent, we are tricked into thinking they are intelligent. This does not mean of course that transformer models - the underlying technology for LLMs is useless. Of course not. But the broadest progress...

35 years down the broad spread personal computing and messaging and clipboard cut/paste has been used increasingly more. The phone apps bubble with their notifications and we shuffle between them moving pieces of information manually from one app to another. It is increasingly difficult to keep track of all the messages on multiple accounts and systems - email, short messages, WhatsApp, Slack, Confluence, Jira, LinkedIn messages, Twitter DMs and so on. Very often the same information is sent using multiple channels. E.g., when boarding a flight I'm getting an SMS, and email and an in-app notification. To be honest an universal messaging client is my biggest hope for the AI. I don't want a self driving car. I want a simple (single!) phone app which would collect, aggregate, and manage the incoming streams of messages, drop spam, unsubscribe subscriptions I never subscribed to, categorize what is urgent and requires attention and what can be handled later and so on. Otherwise we ...

Everything is digital and therefore software-driven today. And that software is getting super complex, with tons of layers and dependencies. This complexity, statistically, leads to many failures which in the end are unavoidable. No matter how thorough the testing is, it is impossible to cover and catch every possible scenario. Even simple stuff like alarm clocks sometimes fail to go off. Something unthinkable in the old days of mechanical alarm clocks. And there are times when a working alarm clock is critical. Such that when you have an intercontinental flight to catch early in the morning. Just recently Apple has confirmed the iPhone alarms may not go off due to a software bug. That is why even being a digital gadget geek, I have a rule to set up an analog alarm clock just in that case - an early morning flight. To regular people this is something close to unthinkable - the most premium mobile phone, with millions of users and billions of R&D and QA budgets behind - can still ...

Economy (as indicated by the GDP) goes hand in hand with energy consumption. The better developed economies consume more and the less developed - well - consume less. And this ratio has been pretty stable over many years, with countries like the United States increasing the GDP and decreasing the energy use slightly. Now with cryptocurrencies and artificial intelligence (AI) in particular, this is probably falling a lot. Both technologies - crypto and AI consume outrageous amounts of energy and contribute very little (if anything) to the GDP. It seems the most progress in AI is in generating noise and misinformation. The Register has reported recently that As The Register opined in 2022 and reported in January this year, search quality has declined because search engines index low-quality AI-generated content and present it in search results. This remains an ongoing area of concern. And this is (unfortunately) in line with my own expectations and experience . I wonder if we are...