JaRail

Cool technique!

I just want to point out that something being science fiction does not mean it's impossible.

(March 2024) Antropic claimed Claude 3 Opus had "graduate-level expert reasoning" with GPQA results of around 60% showing a roughly phd level performance.

(Sept 2024) OpenAI claimed o1 was phd-level in their launch post.

You're kinda both wrong. :)

Yes and no. It would make the test fair but it would also mean losing any optimizations the vendors have made specifically for their model. They're all trained differently so they should be used differently to get maximum benefit. It makes sense to use each vendor's harness to get the most of their models, and compare the results that way. Additionally, it's not easy to use all the models in all the harnesses. Anthropic's subscription only covers claude code for example.

If it was just one test, sure. But if they're spinning these up continuously with new models on tens of thousands of GPUs, air gapping becomes impractical. I would mostly fault them on having no guardrails at all. They should have a monitor/external harness that looks for successful access to external networks then stop it there. They may as well let the models test their own networks for vulnerabilities. That's going to be really important to have going forward.

Has to be mixed. The model accomplished something truly impressive. We'll see how impressive when the zero-days are available look at. But OpenAI as an engineering company screwed up. The impressive part is mostly locked away from public access so I don't see a huge PR upside. The ugly part could bite them and the entire AI industry hard in terms of regulations. People will be citing this for years.

When that happens, I usually just do a manual update on my filters. More often than not, someone has already updated it.

Worth pointing out then that DDG has a number of shortcuts built-in. If you don't care about passing a query through them, it already works.

https://duckduckgo.com/bang

The most-used is probably !g, which sends your query to google. For wikipedia, it's !w.

Even programming questions. If it's something considered popular, you get awful results. Take javascript, I always append "mdn" when I'm trying to look up a language/api detail. Otherwise I'd be sifting through the top ten garbage Q&A or tutorial sites. In the glory days, Page Rank would return reference material first since it was so heavily referenced. But clearly that no longer works in the real world of SEO optimization.

Large governments have access to Windows source code.

https://www.microsoft.com/en-us/securityengineering/gsp

It's more about long-term strategy. China wants full control over their OS. They want features Microsoft would never add. They want support for their own hardware, etc.

Correct. The US only had net neutrality on the books for traditional ISPs. However, they were playing relatively nicely to avoid such regulation happening. The mobile arena was fought over data caps and exemptions rather than throttling.

Data exemptions are absolutely an example of a net neutrality violation. Similar things, like exempting partnered streaming music services from mobile data plans, have been shot down in Canada.

The fix was to block the botnets that were scanning millions of numbers and ban the associated accounts. Likely that includes some ongoing threat detection as well. That'll at least prevent scammers from collecting one more account name/number to attempt exploiting.

It doesn't do anything against a targeted attack against someone who has chosen to be discoverable. That's just how search/discovery is intended to work.

Hiding counts makes it hard to identify imposter accounts and bots. Users need to be able to see account age and counts at a minimum.

No, it wouldn't work. It only works if people can discover you with the "find people you know from your address book" feature. A deleted number won't match. Or you can just turn it off in your discoverability settings.

So people don't generate thousands of accounts then sit on them until they want to spam with legit-looking seven year old accounts.

It's for anti-spam. Preventing people from generating millions of free accounts is valuable.

You can check it all you want. That won't change the physics of its dynamically unstable design.

Exactly. Using Mars as a second example is basically just doubling down on the "Elon isn't trustworthy" argument. That should be a giant red flag to readers.

Have you considered just creating multiple windows/mac accounts? Since you can switch without having to log off, it's really convenient. I understand partners sharing accounts and such but it's really just a better workflow/experience IMHO.

I use Temporary Containers (https://addons.mozilla.org/en-US/firefox/addon/temporary-con...) to do this. You can customize how you want links handled when they target different domains. It takes some getting used to as there are a handful of options to understand/customize. Plus you can have customizations where you're replacing the current container with a new one which kills off your back button history. Just need to be aware of the quirks. It's definitely still in the realm of power-user UX.

Are you sure about that? I don't have Lyft where I live. But, about a year back, I was given a Lyft code for an interview. When I got the code, the instructions made it clear that I needed to register a credit card for liability. I had my account created with my personal credit card and redeemed the code. I thought I was good to go. When I got to the destination city and tried to order a ride at the airport, it said I needed to enter a SECOND credit card for verification. Unfortunately I traveled light and didn't a second card on me.

Tools like this should be built into your IDE. No developer ever wants automated feedback at the end of the process in a code review.

There are lots of academic ML review/suggestion tools. Those people come to the table with trials and statistics to assess the quality of their results. Amazon probably copied one of those papers, added a rules-engine to recommend their own APIs, and slapped a hefty price tag on it.

It'll probably generate irrelevant stats on the engineers to send directly to their managers to use against them in their next review.

> Why don’t we have the ability to restrict at the OS level which domains an app can send information to?

Ads.

Current-gen devices provide an advertising id. It's unique to the device but can be reset to a new random value by the owner in the OS settings.

You can't include Baidu in the list and say Google has 90% market share. Saying Baidu has 1-3% without specifying "in the USA" is misleading. They've got a billion users in China.

I try to use DDG as my default but it's so awful at knowledge-graph questions. I find myself using !g for knowledge queries before even attempting it.

An example would be something like "Rick and Morty episodes." I know google will give me a list with recent/upcoming episode names and air dates for pretty much any show. DDG will link me wikipedia and fan wikis. I make a "<show> episodes" query anytime I want to know when the next episode of something is released.

DDG's knowledge graph (and/or query parsing) is just so limited I skip it anytime I think google will be able to produce the answer directly. Similarly, there are things I'm confident asking a voice assistant, and there are things I won't even bother trying. If it's something I'd ask an assistant, I'm skipping DDG.

Chrome is linked to your google accounts and saves your account auth tokens. When you use incognito, it doesn't immediately send that info. However, it does offer to sign you in. This is different from remembering login/passwords.

I don't think their 10980XE and Threadripper videos from this morning are going to win them any favor with Intel. Absolutely savage rants though.

Intel 10980XE: https://youtu.be/vuaiqcjf0bs

AMD 3970X/3960X: https://youtu.be/a8apEJ5Zt2s