There are no clear details about what DOGE will alter. All the information we have so far amounts to outrage-bait one-liners. Where and how are they deciding to make cuts, and which agencies are they planning to change? Cutting is not intrinsically a good or bad idea, but everything I’ve read about this so far makes DOGE seem anti-intellectual.
> And then I tried gpt-3.5-turbo-instruct. This is a closed OpenAI model, so details are very murky.
How do you know it didn't just write a script that uses a chess engine and then execute the script? That IMO is the easiest explanation.
Also, I looked at the gpt-3.5-turbo-instruct example victory. One side played with 70% accuracy and the other was 77%. IMO that's not on par with 27XX ELO.
Anyone have a technical writeup of the actual bug? I'm trying to explain how this could happen to people who think this is related to AI or cyber attacks.
What happened to the QA testing, staggered rollouts, feature flags, etc.? It's really this easy to cause a boot loop?
To me, BSOD indicates kernel level errors, which I assume Crowdstrike would be able to cause because it has root access due to being a security application. And because it's boot-looping, there's not a way to automatically push out updates?
Doesn’t make much sense for Hikaru to cheat. He competes over the board where it’s much harder to cheat, and he makes most of his money through streaming, not through competition results.
I would love if when buying an iPhone I was able to specify an entire class of apps that I could not install. For example, I could be banned at an OS level from a curated suite of social media apps. They could also make each app report the domains that it connects to so that the ban could continue at a DNS level. Apple is in a great place to do this because of their walled garden App Store model and withholding of root access from users. To undo it you would have to go into a store to use their proprietary software to change the type of apps you're allowed to install.
Pragmatically, we need way more investment in environmental engineering and policies that put conservation first.
Since collectively we can't be bothered because we all need the newest electric vehicle, you can also accept that mass extinction is a byproduct of humanity.
> We document appearance effects in the economics profession. Using unique data on PhD graduates from ten of the top economics departments in the United States we test whether more attractive individuals are more likely to succeed. We find robust evidence that appearance has predictive power for job outcomes and research productivity. Attractive individuals are more likely to study at higher ranked PhD institutions and are more likely to be placed at higher-ranking academic institutions not only for their first job, but also for jobs as many as 15 years after their graduation, even when we control for the ranking of PhD institution and first job. Appearance also predicts the success of research output: while it does not predict the number of papers an individual writes, it predicts the number of citations for a given number of papers, again even when we control for the ranking of the PhD institution and first job. All these effects are robust, statistically significant, and substantial in magnitude.
The paper is paywalled, but that's the abstract.
I'm curious about how much of that is self-sabotage. I don't think we can rule out self-perception of attractiveness as a confounding variable. Internalized attractiveness? If anyone has a good study about that, I'd be very interested.
I'm at about $700/mth but I don't see why I can't make it to $2k monthly. It's a Slackbot [1,2] that assists user story refinement for Jira. If you're a SWE that uses agile methodology this may be interesting to you.
Thank you for commenting these in. I find this quote "That when allowed to be themselves, their psychological issues largely disappear." interesting and hopeful -- and the way it's phrased suggests some causation.
Is there a study that proves this? I did a brief literature scan and it seems to me that researchers are unclear about the causation direction of things like trauma, psychological disorders, and GID/gender dissatisfaction. I think this is a pretty important part of the ongoing debate.
I don't want to be the "citations needed" guy, but I do have a tremendous respect for science and I don't want to just have to go with my gut on something like this.
Oh so there is pre-processing to find the useful portions? What are you using for the pre-processing?
I feel that it's inevitable that OpenAI et al. will be able to handle large PDF documents eventually. But until then I'm sure there's a lot of value of in this kind of pre-processing/chunking.
I'd like legislation that acknowledges both the utility and harm that social media and more generally smartphones have. All I want is an internet-based texting only mobile-phone that doesn't have a browser or anything besides utilities. I don't want any feeds, reels, TikToks, etc. The fact that you basically need a smartphone to socialize, make payments, park, read menus, but it comes with a hamster-wheel for dopamine dispensing is like if the only way you could hydrate yourself is fountain soda -- you're just playing a game where you try not to get diabetes.
Given that the topic is pretty politicized and they didn't go into that much detail about controlling for possible confounding variables, I think the author is trying to express their personal beliefs through a study. I think the data is interesting but the title jumps to a conclusion that IMO the analysis doesn't support (yet).
> Across many applications areas, we’ll be left with a choice of using a 90% accurate model we understand, or 99% accurate model we don’t.
And how do you show that it is 99% accurate besides creating enough automated tests to the point that you could write the procedural version?
I think what I was missing from this article is how to evaluate a domain where neural nets or LLMs can be applied. Image-from-text generation is a great one because accuracy isn't strictly defined. However, telling ChatGPT "code this pacemaker for me" would have a real accuracy attached to it that you could confirm with unit tests.
Had me in the first half, lost me in the second. Why not try to create unions instead of "don't document anything because you're making yourself replaceable"?