Tweet ([Xcancel])(https://xcancel.com/edzitron/status/1809347763349647817) from Ed Zitron: "This report is damning. Goldman Sachs's Head of Global Equity Research doesn't believe that generative AI is capable of solving complex problems, and says the tech industry "is too complacent in its assumptions that costs will decline substantially"", more in the quote tweet.
this post was submitted on 01 Jul 2024
1 points (100.0% liked)
TechTakes
1432 readers
16 users here now
Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.
This is not debate club. Unless it’s amusing debate.
For actually-good tech, you want our NotAwfulTech community
founded 1 year ago
MODERATORS
Expect promptheads to dismiss this as being the ramblings of a "DEI hire" because they're confused by the dual use of "equity" in their title.
Not just promptheads but the entire conspiratorial right is already way into not understanding how these functions and organizations work. There is a weird group of them who thinks that blackrock is a group of woke people (some of them are even saying it is 'THEM') behind everything. Because they confused a talk about 'we need to tell organisations to focus more on diversity' as a pro-diversity stance and not a 'we need to make sure the companies we invest in don't get sued, which causes a drop in share price'.
I’m sure they would find some way¹ to ruin it, but it would be fun if we could convince them to pass a law in some vice-signaling US state that bans private equity’s purchase of every vet and general contractor and empty house.
¹ Anti-semitism, probably.
Posted the Ray Kurzweil thing on Pivot To AI. Let's see how many comments we get that won't be posted.
Behold
Motion capture, but worse - brought to you by AI
(seriously, I saw random Youtubers do better shit than this eight years ago and that is not hyperbole)
wow, that side-by-side is so obviously bad i'm surprised it even got posted. usually AI bros try to at least hide the worst of the tech, or at the very least, say shit like "this is only the beginning!!"
also, was not expecting to click that link and see FUNKe. good nostalgia
it ignores her eyes THE WHOLE TIME!
we finally posted Diz's LLM logic puzzle post to Pivot to AI! Let's see if this draws flocks of excited new users to awful.systems ... we're doing Ray Kurzweil tomorrow, lol.
There was a warcraft 3 pro who played build orders created by Chat-GPT and it was fascinating the degree to which it was able to perfectly imitate the form of the kind of thing you'd find on liquidpedia or some other guide but simultaneously make nonsensical errors that betrayed that it had no awareness. Like, telling you to build a unit of a different race or build without meeting prerequisites.
lol, amazing. and I recall this was one of the areas where people fairly successfully (iirc?) applied GAs to synthesising strategy and build order
(think it was for Starcraft 2 not WC3, but domains very similar)
I've definitely seen some impressive machine learning-based outcomes (definitely fucking up the technical details here) but there's a world of difference between a system trained to play StarCraft and a system trained to predict the next bit of text.
oh absolutely - part of what's entertaining is that the dipshits don't even seem to realize that (or don't admit it?)
the fuck
the only other thing on there is also confusing as fuck:
”Mirrors” is one of our services that developers and non-developers could use on their websites or in their favorite web browsers. It can reduce server load by acting as an extra caching layer for every direct link available on your website.
All data will be cached for 86400 seconds or 1.00 day(s), and distributed globally, effectively increasing performance and reliability, which will lead to an improved user experience.
so, ah, essentially a demo web app supposedly running at CDN scale, and supposedly powered by AI (to do what? this is not answered, of course). this is almost definitely just some basic cloud shit someone spun up, but like, why? there’s money in it if all this shit is logging DNS queries and web site accesses to sell to advertisers or for other much more nefarious purposes (or they’re doing injection on high value targets), but surely people who change their DNS settings and developers aren’t stupid enough to fall for this just cause it says it’s AI… right?
Q: Is this really free?
A: If it's not, you should already see a Pricing page.
this really is the slimiest way to avoid saying “it’s free until we say it isn’t”
wonder if it's white-label Cloudflare
I got curious and looked up both IPs associated with the domain name and it is in fact just CloudFlare
also, their DNS-over-HTTPS (this is new to me but I stay the fuck away from cloudflare and it looks like a standard they’ve pushed for) endpoint is just the 1.1.1.1 one with the domain name changed, but the AdGuard and Accelerator ones basically confirm they’re analyzing traffic before passing it off to cloudflare
a quick check using a canary. test command:
curl -H 'accept: application/dns-json' 'https://0ms.dev/dns-query?name={snip}.canarytokens.com'
canary trigger:
geo_info: {'loc': '51.5085,-0.1257', 'org': 'AS13335 Cloudflare, Inc.', 'city': 'London', 'country': 'GB', 'region': 'England', 'ip': '141.101.70.74', 'timezone': 'Europe/London', 'postal': 'E1W', 'asn': {'route': '141.101.70.0/24', 'type': 'hosting', 'asn': 'AS13335', 'domain': 'cloudflare.com', 'name': 'Cloudflare, Inc.'}}
so, yeah, the actual resolve is being done by cloudflare servers too - it's not even just a cf frontproxy to a different backend service. could be done with cf workers or something, I imagine, would need to test a bit further to know/try see
that's fabulously odious!
yeah, think i'll be writing this one up for the weekend
this sort of thing does not help me have an anti-carceral attitude
This is slightly weird. There's a company called "DNSfilter" that's been advertising itself for a few years as an "AI-powered DNS resolver" - since well before the present hype. I suspect their "AI" bit is threat detection with machine learning in. They appear to be going for the lucrative boring enterprise market.
This thread suspects the whole 0ms.dev thing was written with LLMs and the "mirrors" bit is basically an open proxy.
You sure they're skimming results? I saw what looks like a filter to add to an adblocker ...
and the “mirrors” bit is basically an open proxy
could test this by setting up a simple example page and seeing the originating source for the traffic
same technique as previous, using an http (not https) canary:
geo_info: {'loc': '51.5085,-0.1257', 'org': 'AS13335 Cloudflare, Inc.', 'city': 'London', 'country': 'GB', 'region': 'England', 'ip': '141.101.98.196', 'timezone': 'Europe/London', 'postal': 'E1W', 'asn': {'route': '141.101.98.0/24', 'type': 'hosting', 'asn': 'AS13335', 'domain': 'cloudflare.com', 'name': 'Cloudflare, Inc.'}}
useragent: (no user-agent specified)
request_headers: {'Host': 'canarytokens.com', 'X-Real-Ip': '141.101.98.196', 'X-Forwarded-For': '2a06:98c0:3600::103, 141.101.98.196', 'X-Forwarded-Host': 'canarytokens.org', 'Connection': 'close', 'Accept-Encoding': 'gzip', 'Cf-Ray': '89f1a3a114752508-LHR', 'X-Forwarded-Proto': 'https', 'Cf-Visitor': '{"scheme":"https"}', 'Cf-Ew-Via': '15', 'Cdn-Loop': 'cloudflare; subreqs=1', 'Cf-Connecting-Ip': '2a06:98c0:3600::103'}
request_args: {}
idle thought: since this is (as @self highlighted) effectively an open proxy, you could also just resource-bomb it quite easily
not that I'm planning to, but it'd be dirt bloody easy and it would be a very quick test as to the claims on the site
while looking into this I was surprised to find that hickory doesn't seem to have a cli binary in its crates. did find crab-hole and adguardian in the process though, gonna have to play with those later
You sure they’re skimming results? I saw what looks like a filter to add to an adblocker …
see I don’t have a smoking gun on this, but if it is a garden variety adblocker and whatever the accelerator does (it’s extremely unclear) running on cloudflare workers, they at least have the ability to MiTM and modify DNS queries, and open proxies are always a security nightmare. my impression is the filter list they link is either what that endpoint uses to filter, or it’s LLM vomit they kept in because it felt more legitimate.
this might be worth poking at more in a controlled setting — a command line DNS-over-HTTP client using their endpoints and some exploration of the open proxy with curl (especially with sites that utilize credentials, with great care taken to use disposable accounts for this) might be illuminating
view more: next ›