Things that caught my attention
My friend and data guru Edafe wrote a brilliant commentary to last Quantum's link drawing similarities between transformers and compressors. Her take is on the relationship between LLMs and neurodivergent brains and it resonated a lot. Edafe maps her own AuDHD brain (the "autistic wolf" and the "ADHD wolf") onto how LLMs work: both are pattern-matching, context-hungry, limit-bound "fancy autocomplete" running on a very greedy battery. She's blunt about the trade: she uses LLMs as magnifiers and translators of unspoken neurotypical rules, but never to outsource thinking, because unguarded use breeds "domestication through convenience." You'll find it in the AI section.
Similarly mind-blowing is the tutorial showing that executables in the ELF format are fundamentally databases, and shows how to interact with them with SQLite. As the author says, "ELF is a database that refuses to admit it". Go to the tools & tutorials section.
Also, note the two topical data visualisations by the European Correspondent. They are becoming truly a brilliant source of insightful dataviz, as is Data Sheets with their recent exploration of the Butterfly Effect.
Worthy of attention, but more on the geeky side, the brilliant story about training a 125M-parameter Model to Autocomplete Piano. Listen to the autocompletion of Beethoven's Für Elise, and tell me it's not amazing 😃
·Wired has a good story about how "easy" it has become to apply to jobs these days. "Thanks to a dwindling supply of open roles, “one-click” applications, and the rise of artificial intelligence, it’s easier than ever to apply for a job. We’re all paying the price."
There are multiple factors at play here, but AI is becoming the most prominent. In my previous job, I tried to recruit our Head of Product Management. Previous experience in hiring at similar level: no more than 30-40 applications, if we were really lucky. This time last year, we had 300. A whole lot of AI slop, and aside from the obvious time wasted, I began to wonder what that would do to the accuracy of our (brilliant, but very human) shortlisting panel. An obvious reaction to this – which I've discussed in previous issues of Quantum – is to throw AI at this problem to help the struggling panels. But this introduces questions of bias and fairness (and, yes, as I wrote in QoS 672: humans have these traits too).
Recruitment is just one area where AI's contribution is complicating things and muddling ethics. The BBC reports that schools and councils are now struggling with fully AI-assisted complaints, adding strain because of an increase in both the number and the length of those complaints. Services like this are actually advertising themselves as adopting AI to automate processes like the railway's Delay Repay form-filling.
The problem that's emerging here is that many of our societies processes were designed with a reasonable level of friction for humans, and this required a certain level of resources. But AI is not subject to friction the same way, and therefore the process' resourcing assumptions no longer hold true. When friction stops in our AI-driven world, will our systems fall apart?
·Quantum #669 had an open rate of 44% and a click rate of 8%, and the most clicked link was this data-driven look at link rot on the old web.
'till next week,
Giuseppe @puntofisso.bsky.social
