Using natural language prompting almost certainly makes LLMs a) need to be larger than they could be, b) less efficient and reliable than they could be, and c) much harder to use than they could be.
@fsg
Software Developer, mate & meme connoisseur๐ง๐จ๐ฝโ๐ป Born in Argentina, living in Italy | Master in Mass Media & Politics | former translator and journalist now writing mostly code | Staff at @journalismfestival.com Writing in English ๐ฌ๐ง, Spanish ๐ฆ๐ท and Italian ๐ฎ๐น
Using natural language prompting almost certainly makes LLMs a) need to be larger than they could be, b) less efficient and reliable than they could be, and c) much harder to use than they could be.
This is going to be the big โfind out.โ The hard part of producing good software that does useful things is getting real clarity on what those useful things are. Coding forces dimension reduction because its language is so primitive. Without that back pressure, we get slop.
That's why in order to use an LLM with success you need to apply the same diligence you have for programming, and on top of that you should find specific terms and avoid ambiguous ones (something a programming language already does for you).
Think about it: for writing instructions it's better to have fewer words and verbs at your disposal, that's what makes coding capable of being deterministic, in other words that's programming. And it's not going away with LLM's, it's actually more tedious to avoid natural language humongous choices
Having a linguistics background I think there's a huge misconception when it comes to prompting LLM's, the vast majority thinks that coding is "hard" compared with speaking a natural language. Well it's actually the contrary from a utilitarian POV less verbs at your disposal => easier to be specific
Itโs so funny to think about the fact that thereโs people out there deferring every life decision to this
unfortunately this is already happening :/ and the worst part is that people is learning arbitrary class names instead of standard property names. It's Bootstrap all over again (but this time atomic and specific, at least bootstrap have few general semantic coherent class names)
Paris prosecutors raid the French offices of Elon Muskโs X as part of an investigation into spreading child pornography and deepfakes.
Really insightful talk ๐ซถ๐ฝ
#FOSDEM @alex.bsky.team
Presenter at front of lecture hall demonstrating an open source docs interface projected on a large screen while audience members take notes and photos.
Why the French govt digital services chose #Python and #Django: โthe boring stack can be upgraded easily and will still work in 20 yearsโ -- Manuel Raynaud showcases LaSuite #FOSDEM. Open source digital collaboration apps, built as an alternative to Microsoft Teams / Office
#FOSDEM is basically a Linux Day on steroids distributed all around the city ๐คฉ
A beer can branded FOSDEM
#FOSDEM
A great piece by Donald MacKenzie on the past, present and future of Google, the most powerful tech giant and perhaps the most vulnerable too. Insightful and well written
www.lrb.co.uk/the-paper/v4...
Those who can code, code.
Those who canโt code, teach coding.
Those who canโt teach, post on LinkedIn. ๐
This only happens to you once
NPM's week
Ciencia en la era del 'streaming': el fenรณmeno viral de la expediciรณn a 3.900 metros de profundidad en el Atlรกntico Sur
www.eldiario.es/ballenablanc...
How many times do we have to remind people: the chatbot *cannot* admit to *anything*. It *doesn't know*. When you ask it about itself, it generates its answer the same way it generates your request that it write an email. It takes what you tell it and responds in way to sound plausible.
"Being able to put a name to something abstract allows you to more easily build an argument about it [...] Donโt let inevitabilism frame the argument and take away your choice."
โThe Danish government is to clamp down on the creation and dissemination of AI-generated deepfakes by changing copyright law to ensure that everybody has the right to their own body, facial features and voiceโ
www.theguardian.com/technology/2...
Unlock AI for 50% off Slack's Pro plan now includes AI features. Upgrade by July 17 for 50% off your first 3 months. Learn more Instantly summarize channels and threads Unlock unlimited message history Work with people outside your organization
I am once again asking for some equivalent to a pop-up blocker that blocks the "would you like AI with that" messages sprouting up like spots of mold in every application
The superintelligent AI is smoke and mirrors. Big corps and states with do atrocious things to humans and blame the AI god. So many will go towards religious psychosis and be radicalized towards fascism. Instead of AI, it's the billionaires benefiting from all this we should be looking at.
*Steve Bannon this whole time*
Il โdecreto sicurezzaโ รจ lโennesimo tassello dellโorbanizzazione dellโItalia.
Norme ad hoc contro i nemici politici, crudeltร verso i piรน deboli e repressione del dissenso.
La deriva democratica meloniana prosegue sempre piรน veloce e inesorabile.
Ne ho scritto su @valigiablu.it ๐
Dear @bsky.app devs - I am continually confused by the "Share" button that looks like an "Upload" button.
Please consider using the icon that looks like interconnected nodes, or the swoopy arrow thing.
Thank you!
Josรฉ Mujica, the charismatic former guerrilla fighter who later went on to lead Uruguay and became known as "the world's poorest president" for his austere lifestyle, has died at 89.