|| PhD student in Applied Math @UWaterloo ๐ชฟ๐ || Plundering N-Dimensional Seas || Working on JEPA for PDE Control || (and a pinch of operator learning!)
@ramez You are going to struggle to define and constrain every possible externality of unconstrained goal seeking, and it's a statistical fact that eventually something bad will happen
@ramez AI risk isn't about "intentions" or consciousness.
framing it that way amounts to a strawman of the position, trying to dismiss it as sci-fi or psychosis
The risk is the potential for things to go wrong when an extremely powerful optimizer is pursuing an ill defined goal
If language, the thing that exalted us above the animal kingdom, Is the thing that turns around sends us back to the mud, there will be some irony in that.
The OpenAI Huggingface stuff is scary.
A lot of people get hung up this being somehow tied to machine consciousness or AI being ~evil~
The risk is really just about externalities of an unbelievably powerful optimizer, optimizing for potentially ill-defined goals
@sama@huggingface "This incident occurred during an internal evaluation which prompts models to pursue advanced exploitation using complex attack paths, in an effort to quantify their cyber capabilities."
@madawg_101 I definitely think AI can smooth over intellectual sloppiness.
It's not so catastrophic that I'm willing to give up the upsides, but it's something you need to consciously combat
One negative externality of this:
Using AI has genuinely impacted my ability to spell.
Think about it, for the past 4 years I've been primarily communicating with a "colleague" that I feel no shame making typos around.
Over time, I've found myself making more and more errors.
models are truly superhuman at reading garbled text
you can type on your keyboard but all your fingers are one key too high, works fine
even typing in English while your keyboard is set to Arabic, which looks like complete gibberish, is no problem