Reading through Infinite Jest by David Foster #Wallace, I wonder if we are ready for a new-new sincerity period. On the one hand, themes from the novel still feel relevant, so maybe we've never been able to escape irony as escapism.
I'm only half way but there is a sense of urgency in the subtext that I can't really articulate.
Every time an #AI company says they release a new model, I wonder whether this is actually true or if they just daisy-chain existing models, increase compute costs and package that as a new 'product'.
After awhile, costs are shifted over to the user and the cycle repeats. But surely at some point this has to end. It just isn't sustainable in an economic or environmental sense.
Wat gebruikers van een #LLM moeten begrijpen, is dat deze systemen geen waarheid kent.
Dat maakt niet uit wanneer je het als programmeerhulp gebruikt, want een werkend stuk code is niet 'waar', hoogstens zou je kunnen zeggen dat het werkt of niet.
Maar voor het genereren van natuurlijke tekst is dit wel een probleem, want zelf was de tekst de werkelijkheid weerspiegeld, is dat bij toeval
Dit verschil niet onderkennen is post-truth, waarbij het belangrijker is dat het waar zou KUNNEN zijn.
There seem to be two types of nerds. Those who want to deconstruct everything to understand the very core of the technology they are using, and those that vibe and aren't inherently interested in how anything functions and with what consequences.
The first type are the true geeks, the second are zealots waiting for someone to give them all the answers.
Sadly it's the second type that are celebrated as being part of the way forward.
Recently my partner passed away and now I feel more than ever the lure of an #llm . I mean, won't be interesting - for research purposes - to train a local model on his chat history?
Of course not! It would be a simulacrum of his speech, only consist of a shiver of it because most of our conversations haven't been through a messaging app. It would be my feelings, remixed by a language model.
In this time of vulnerability I can see how many people fall for this trap.
Ben je bezig met #LoRa? Welk protocol gebruik je?
How well do relatively small open weight #LLM's hold up to their closed, paid services? I'd never use one of those walled gardens, but running Mistral Small (28b) locally is pretty interesting so far.
Though to say I'm impressed is pushing it. Anyone know of comparisons?
May be more relevant now that #OpenAI and the like started to (silently) route users to less powerful models in their services. So very soon, performance of paid services may be comparable to the open weight ones?
Anyone got #readingrecommendations on #Capitalism and #degrowth? I see books out now that call for #deregulation, that #Technology will be able to give #abundance.
At the same time, #exploitation and #colonialism seem to be in the middle of this push.
Wonder if there already are sources exploring this.
(Note to self: stop to procrastinate by wanting more sources and start writing...)
With #LLM's being misused all the time, its like how I imagine how it was when cars were becoming commonplace but there were no official or implicit rules yet.
Everyone was probably racing around, being sure of their own ability to drive well, but these should be rules for others!
Will there be rules on how to not use AI? Who knows. The downsides are less visceral than car crashes.
Het lijkt erop dat de grote afkeer tegen Trump, Musk, Vance etc nu helpt om degelijke partijen aan de macht te helpen.
Heel mooi, want het leek er toch een beetje op dat de kikker in een opwarmende pan metafoor waarheid leek te worden.
Het tij is nog niet gekeerd. Verre van. Maar het biedt hoop.