dwaltrip

32 ms·

571.
by dwaltrip 3y ago
People are usually a lot more understanding and less judgmental than we imagine. Especially if we act in a way where we don’t need them to understand. Where we remain true to ourselves and find our inner security. If someone does judge yo
572.
by dwaltrip 3y ago
Gpt-3.5 or 4?
573.
by dwaltrip 3y ago
Instead of something extricable, I imagine a series of processes that, due to specific conditions, results in some new process that has distinctly different characteristics from the originating process. E.g. snowpack melt -> mountain str
574.
by dwaltrip 3y ago
Note, Gemini Ultra, which they claim is competitive with or possibly even better than GPT-4, isn’t out yet. They have released a weaker model, Gemini Pro. It will be interesting to see how capable Gemini Ultra actually is. For now we wait.
575.
by dwaltrip 3y ago
Adding on to this, companies as “AI-orgs-running-via-human-individuals” can be checked by other similar entities of comparable competency, e.g. activist groups, governments, non-profits, and so on.
576.
by dwaltrip 3y ago
The solution I’ve found is to try to optimize less. And to develop trust that your future self will be able to handle whatever comes. And to realize that shit happens and that’s okay too.
577.
by dwaltrip 3y ago
Thinking this through out loud... I think there's a difference between trying on the clothes in the store vs. buying online, trying on, and returning. I bet the retailer has much lower confidence that the clothes weren't compromis
578.
by dwaltrip 3y ago
Sure, that may be. But you still have to ride the first half of the curve, before it inflects. I'd rather make sure the ride isn't too bumpy.
579.
by dwaltrip 3y ago
Something seems off. Look at the website in their HN profile and their LinkedIn. They may need help. Bion, if you are reading this, I hope you are doing okay buddy. It’s never too late to ask for help. As someone who has been through the ab
580.
by dwaltrip 3y ago
+1 for the Teixcalaan series
581.
by dwaltrip 3y ago
> Some reasons we might expect this to happen are: 1. People don't want to win or don't care about winning 2. People understand their mistakes but haven't put in enough time to fix them 3. People are untalented 4. People d
582.
by dwaltrip 3y ago
Couldn’t we architect or train the models to differentiate between streams of input? It’s a current design choice for all tokens to be the same. Think of humans. Any sensory input we receive is continuously and automatically contextualized
583.
by dwaltrip 3y ago
If the info is in the system prompt, you can trust it. You'd think they would indiciate if it's Gemini in the system prompt.
584.
by dwaltrip 3y ago
I don’t think that’s the case.
585.
by dwaltrip 3y ago
You are right, prediction doesn’t provide one with base motivations or drives. But those seem fundamentally different than raw intelligence. Your examples aren’t an issue I don’t think. To take one: efforts to minimize one’s consumption to
586.
by dwaltrip 3y ago
When in comes to training LLMs, the definition of “improvement” is incredibly clear, as one must literally code a loss function that the model then minimizes. It gets murkier trying to map that actual capabilities, but so far, lower loss ha
587.
by dwaltrip 3y ago
Hypothesis: Intelligence is prediction?
588.
by dwaltrip 3y ago
You bring up good points! For what it's worth, I wasn't arguing against livable wages. I simply wanted to point out the error in the parent comment's argument.
589.
by dwaltrip 3y ago
You're just kicking the can down the road. Where does the "god entity" come from? It's the exact same question as "why did the universe have such and such initial conditions?". We've added a new variable,
590.
by dwaltrip 3y ago
> The absolute floor for wages should at least meet a person's non-discretionary expenses plus a bit of discretionary spending if they're working close to full time. If it doesn't meet that minimum level, then the employee
591.
by dwaltrip 3y ago
Are you still working at the cafe? How do you like it?
592.
by dwaltrip 3y ago
What reason did they give for firing you?
593.
by dwaltrip 3y ago
Try Andrej Karpathy’s zero to hero course. It’s very good. It’s 8 video lectures where you follow along in your own Jupyter notebook. Each lecture is 1-2 hours.
594.
by dwaltrip 3y ago
Apparently there are roughly 2 trillion text messages sent per year in the US [1]. I did a sanity check, that’s like 40 or so a day per person, so sounds reasonable. I couldn’t find the average message length, but I would guess it’s fairly
595.
by dwaltrip 3y ago
Holy shit, you are right. They probably have 10-100x the data used to train gpt-4. Decades of every text message, phone call transcript, and so on. I can’t believe I haven’t seen anyone mention that yet. People keep saying we don’t have eno
596.
by dwaltrip 3y ago
For the current generative AI wave, this is how I understand it: 1. The scaling path is decreased val/test loss during training. 2. We have seen multiples times that large decreases in this loss have resulted in very impressive improve
597.
by dwaltrip 3y ago
Is there any major news reporting on the Annie Altman stuff? That looks like front page material to me.
598.
by dwaltrip 3y ago
You may be ramping up too quickly. It doesn’t need to feel like torture. Drop any preconceived notions of what counts as an “exercise session”. Go as slow as you need. Let your body adapt. When I first got back into running 2 years ago afte
599.
by dwaltrip 3y ago
LLMs predict language, and language is a representation of human concepts about the world. Thus, these models are constructing, piece by piece, conceptual chains about the world. As they learn to construct better and more coherent conceptua
600.
by dwaltrip 3y ago
What I found most interesting were the comments at the end. Altman states that he expects next year’s improvements to the capabilities of leading models to take everyone by surprise. Of course, this was right before the recent chaos… ——— Jo

More ›