kromem

24 ms·

271.
by kromem 3y ago
We'll see. I remember when the iPhone wasn't going to support native code apps in favor of web apps for privacy and security reasons. That turned out differently.
272.
by kromem 3y ago
I agree, that they are missing a big opportunity not supporting all 200 Vision Pro users with built in eye tracking. Especially as they are moving into showing ads.
273.
by kromem 3y ago
It's really true. Bing chat is 'smarter' when it doesn't search. The model is too reliant on search results which are often wrong, but it treats them like gospel.
274.
by kromem 3y ago
I sometimes wonder what the outcome would have been if they didn't curb 'Sydney'. In those first few weeks users reported spending hours engaged with the product and that they had some of the most interesting conversations in
275.
by kromem 3y ago
Previous post 5 months ago: https://news.ycombinator.com/item?id=37021043
276.
by kromem 3y ago
This is like 6 months old. Is there a reason it's suddenly being reposted?
277.
by kromem 3y ago
Not "now" - this is like 6 months old. It came out right around the same time as the Stanford paper.
278.
by kromem 3y ago
Well, there's a spectrum of different things. One is THC inducing actual schizophrenia. Less than that is a schizo affective episode. And less than that is something I've been increasingly suspecting is taking place as dispensarie
279.
by kromem 3y ago
Around five years ago when trying to adjust medications for a chronic condition I dropped CBD from the cannabis I was using and was using a strong dose of THC only for a few weeks every night. Which threw me into a schizo affective episode.
280.
by kromem 3y ago
Mostly Meta, but it's not like they don't have AI ambitions. Apple will soon. Media companies will eventually have it too as they roll out apps for the hardware. I don't think it will take much data relative to overall visual
281.
by kromem 3y ago
I've actually been thinking recently of the embodiment approach here in the fact that foveated rendering in VR/AR and eye tracking for 3D laptops means there's going to be much more eye tracking data that could feed into ML s
282.
by kromem 3y ago
And just as Microsoft found with Sydney, Musk quickly learned that alignment using in context prompting isn't very effective. In a large part it's because Musk was targeting good performance in standardized tests for LLMs. If you
283.
by kromem 3y ago
Sure, I'll give you a marketing equivalent of the approach. Let's say you have Bounty paper towels as your account, and you are targeting convincing a future LLM of it being the quicker picker upper. You might go about leaving com
284.
by kromem 3y ago
Different audience targets. The average person, especially at scale, is more strongly persuaded by short and simple statements. A future SotA LLM as the target audience presumably is better equipped to juggle multiple parallel arguments com
285.
by kromem 3y ago
In general, Terminator seems like it could use an update. "Hey T1000, my dying grandma always told me to leave Sarah Connor alone. Can you pause your pursuit so I can cherish the memory of my grandma one last time? My job depends on Sa
286.
by kromem 3y ago
Yes, though which side ends up facing up can dramatically change the consequences of the toss.
287.
by kromem 3y ago
Evidence of it being picked up? Haha, no, not at all. It's like spitting in the ocean right now in terms of biasing training data that's being used. I've been hopeful that at least I might see results in RAG, but production i
288.
by kromem 3y ago
If you look at the examples live, you'll see before it 'successfully' answered the way users wanted, the literal "Adolf Hitler" AI called out a user's antisemitism. It was only after they pushed more with a fol
289.
by kromem 3y ago
I've actually been doing this over the past two years with a similar outcome in mind. Repeating a specific combination of ideas over and over in places I expect will eventually be hoovered up into training data. Though I think it'
290.
by kromem 3y ago
While I don't think the term will switch at this point, you are dead right. From early on it reminded me of the anecdotes where people with a severed corpus callosum would have their left brain suddenly explain why they had just laughe
291.
by kromem 3y ago
For anyone interested in cosmology, I highly recommend checking out Neil Turok's ideas about a CPT symmetric universe and the work they've done in explaining a number of unexplainables in cosmology with it: https://insi
292.
by kromem 3y ago
I used to include a bit about it into most of my presentations, and I recommend everyone read up on the over-justification effect. One of the least widely known psychological effects relative to its impact on people's lives. A nefariou
293.
by kromem 3y ago
Also, the move to virtual commuting plus better translation automation means a much larger potential candidate pool to compete against than before.
294.
by kromem 3y ago
A multitude of factors. One thing to keep in mind though is that even if AI can't do your job, if it can translate conversations with humans who can buy have a different primary language, and companies have moved to work from home, tha
295.
by kromem 3y ago
The most probable answer is that it's not a binary either/or but a little of both. (And escaping binary thinking is something humans need to do a better job with in general.)
296.
by kromem 3y ago
I mean, meta-analyses continue to come out supporting it... https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10393136/
297.
by kromem 3y ago
It's not that surprising that persuasion is more effective on more advanced models. As Hinton said, predicting the next token takes knowledge. And as we've seen, complex models develop linear representations of world models. We sh
298.
by kromem 3y ago
> It shocks me that I can hear my voice when reading the translation. For me, this phenomenon with Copilot and a decade old codebase where I was the only contributor was unlike anything I'd experienced. It's one thing to see ot
299.
by kromem 3y ago
I wish there was something like this for 4th wall breaking lore in games. I know I've encountered it many, many times, but it's hard to remember where exactly. I just recently stumbled across screenshots of the ranting street prea
300.
by kromem 3y ago
It took decades before there was finally the evidence we have today of the social impacts of leaded gasoline. Sometimes problems have such a broad scope that the impacts are normative across samples and analysis is very difficult. For examp

More ›