This is very interesting research. This confirms that LLMs can't be trusted on any output they make about their own inference. The example about simple maths is particularly striking, the real inference and what it outputs if you ask about its inference process are completely different.
Now for the topic dearest to my heart: It looks like there's some form of concept graph hiding in there which is reapplied across languages. Now we don't know if a particular language influences that graph. I don't expect the current research to explore this question yet, but looking forward to someone tackling it.
Unsurprisingly, hiring scams are becoming more elaborate. Keep it in mind for your upcoming interviews.
The "asleep at the wheel" effect is real with such tools. The consequences can be dire in quite a few fields. Here is a good illustration with OSINT.
Even if you use LLMs, make sure you don't depend on them in your workflows. Friction can indeed have value. Also if you're a junior you should probably seldom use them, build your skill and knowledge first... otherwise you'll forever be a beginner and that will bite you hard.
We just can't leave the topic of how the big model makers are building their training corpus unaddressed. This is both an ethics and economics problem. The creators of the content used to train such large models should be compensated in a way.
Between this, the crawlers they use and the ecological footprint of the data centers, there are so many negative externalities to those systems that law makers should have cease the topic a while ago. The paradox is that if nothing is done about it, the reckless behavior of the model makers will end up hurting them as well.
Unsurprisingly, Wikimedia is also badly impacted by the LLM crawlers... That puts access to curated knowledge at risk if the trend continues.
I somehow recognise myself in this piece. Not completely though, I disagree with some of the points... but we share some baggage so I recognize another fellow.
Sure, a filter which turns pictures into something with the Ghibli style looks cute. But make no mistake, it has utter political motives. They need a distraction from their problems and it's yet another way to breach a boundary. Unfortunately I expect people will comply and use the feature with enthusiasm...
Again that confirms that all the hype and grand announcements are not deserved. It also gives a good idea of the skills which are required to use those tools, clearly the setup process is involved if you want to don't want to be overwhelmed and drowning in bad code.
This is definitely an interesting declarative language. Looking forward to more such neurosymbolic approaches.
And yet another reverse proxy to use as a scraper deterrent... It looks like several are popping every week lately.
Don't underestimate how much of a skill making a stupid crawler can be...
When a big player has to prepare a labyrinth of AI generated content to trap bots used to feed generative AI learning pipelines... something feels wrong.
Despite the marketing speak... it's definitely not there yet. So far all the attempts at using LLM for coding larger pieces end up in this kind of messy results. It helps kickstarting a project indeed but quickly degenerates after that.
More details about the impacts of the LLM companies acting like vandals... This is clearly widespread and generating work for everyone for nothing.
Impressive new models for text to speech.
Still so reliable... could we confine this to NLP uses please? Should never had been used for anything looking remotely like search.
Those bots are really becoming the scourge of the Internet... Is it really necessary to DDoS every forge out there to build LLMs? And that's not even counting all the other externalities, the end of the article make it clear: "If blasting CO2 into the air and ruining all of our freshwater and traumatizing cheap laborers and making every sysadmin you know miserable and ripping off code and books and art at scale and ruining our fucking democracy isn’t enough for you to leave this shit alone, what is?"
Words are important, I'm dismayed at how the marketing speak around generative AI is what people use... that completely muddies the thinking around them.
More smaller footprint models are becoming available. This is becoming interesting.