HNHacker News
TopNewBestAskShowJobs

tallytarik

327 karma · joined April 19, 2017

Canberra, Australia.
submissionscomments
tallytarik··on IPinfo started offering free unlimited country-level geolocation and ASN details
This is an interesting point and in my view, there is no clear answer here.

Here's an example: https://ipinfo.io/172.224.238.32

This IP is actually in Ontario, which you can easily verify with a ping measurement. But it is announced as being in Calgary by Apple's iCloud Relay geofeed (https://mask-api.icloud.com/egress-ip-ranges.csv).

Why? Because the intent of iCloud Relay is to obscure a user's IP address while still providing a roughly accurate location, specifically so that geolocation-based services still work as expected. For that to work, they need to provide 'fake data' in this geofeed so that they have pools of addresses covering thousands of cities around the world, AND they need geolocation providers to accept this.

ipinfo accepts this, even though it's wrong. So do we. After all, geofeeds were supposed to provide a public geolocation database, the idea being that the network operator should be trusted to have the best information. We could provide the 'real' location, but if 9 providers say an IP address is in X and 1 provider says it's in Y, and Y is correct, you may just be frustrating end users of the network.

But where is the line? I'm not sure, and it's hard to say who has the balance right here.

We try to mitigate this by providing extra data like whether the address is a hosting or relay provider - for free, unlike others :) Some future addition could be to provide additional accuracy or source information, or even a 'reported' vs 'measured' location.

We're working through this and hope to get to the right answer over time. Thanks for raising this. :)

tallytarik··on IPinfo started offering free unlimited country-level geolocation and ASN details
We and many others use the same techniques too. This is a concept introduced 20 years ago, and RIPE Atlas/IPmap has been a public implementation of this idea for the last decade or so.

(However, the vast majority of IPs can't be geolocated in this way, and there are caveats to those that can be.)

In any case, the difficulty with all providers in this space is how you prove accuracy at scale. If we assume some provider has some proprietary technique that nets 100% accuracy, that's great, but what do you compare it to? There is no ground truth data source - we are supposed to be that.

Marketing plays a big role, and admittedly, these guys have much better marketing on this point :)

tallytarik··on IPinfo started offering free unlimited country-level geolocation and ASN details
If you need more than country level data, these folks become very expensive very quickly.

Many providers are orders of magnitude cheaper. I run one of them - https://iplocate.io - but there are plenty of other high quality and affordable services.

tallytarik··on You wouldn't steal a font
I suspect they are printing more money with the 1-10 megacorps who can afford to pay millions of dollars for per-eyeballs licenses.
tallytarik··on Geocoding APIs compared: Pricing, free tiers and terms of use
IPLocate.io is free to use up to 1,000 reqs/day, also includes a bunch of other IP data like proxy/vpn detection, asn/company/org info. Disclaimer: I run this service, so I'm biased :)
tallytarik··on Default styles for h1 elements are changing
The warning is only for <h1>s inside a section, so GP’s example — where the only h1 is at the top level — would not be flagged.
tallytarik··on Thank HN: The puzzle game I posted here 6 weeks ago got licensed by The Atlantic
I see that as the core point of the game! Would it be a whole lot of fun to complete the game in one guess?

On the other hand I like when I can ‘skip ahead’ and then try to reverse engineer a trickier inside clue.

tallytarik··on Apple's AI isn't a letdown. AI is the letdown
“Hey Siri open the curtains”

“I found some web results. I can show them if you ask again from your iPhone”

Nah, Apple is the letdown, and has been since before ChatGPT.

tallytarik··on Mountains, Cliffs, and Caves: A Guide to Using Perlin Noise for Procedural Gen
The article mentions Minecraft — if you’re interested, check out “Reinventing Minecraft world generation by Henrik Kniberg”, a great talk about procedural generation by one of the developers: https://youtu.be/ob3VwY4JyzE
tallytarik··on Apple's Software Quality Crisis
Notes is so bad. Often copy and paste just breaks. It will copy some string of characters from a different part of the note than you selected, until you restart.

Reminders too. For at least 5 years, creating multiple reminders in a certain order will not create multiple reminders, and will just append the text to the end of the first one. Surprise — hope you didn’t set any important reminders in that batch!

And the great thing about their standardization between iOS and macOS code means the exact same bugs exist in both versions. Yay.

tallytarik··on Decompilation of Minecraft: Legacy Console Edition
I remember scoring an early key for the Xbox 360 edition by solving a puzzle tweeted out by one of the developers. I was thrilled… my friends weren’t!
tallytarik··on Voice Isolator: Strip background noise for film, podcast, interview production
I know the sorts of “creators” you’re talking about, but I’ve never heard of this as a response before.

Are there really that many people who 1) are aware that this could be effective, and 2) are quick witted enough to pull their phone out and play music in response to being harassed?

tallytarik··on OpenAI didn’t copy Scarlett Johansson’s voice for ChatGPT, records show
Who slandered who?
tallytarik··on Apple announces new accessibility features, including eye tracking
> “We believe deeply in the transformative power of innovation to enrich lives,” said Tim Cook, Apple’s CEO.

Does this line actually mean anything? Press releases are so weird.

tallytarik··on macOS Sonoma is setting records for update size
Many here seem to have taken the wrong implication from the headline and graph.

This article is saying Sonoma has the smallest updates in years, and that installation time is decreasing.

tallytarik··on macOS Sonoma is setting records for update size
The article actually says the latest updates are the smallest in years.
tallytarik··on Seraph Secure: anti-scam software co-founded by Kitboga
Kitboga has shown in his videos where a victim will be at the store, and will either see a sign that warns about gift card scams, or is specifically told by the cashier, and still buys them.

Why? Because often the scammer already has remote access to their computer and has threatened to lock it permanently, or take over their bank account, or whatever else. Maybe they’ve already used the infamous Inspect Element to ‘zero’ their bank account. Scam or not, the victim believes that they’ve already lost.

You would need a cashier to have, essentially, Kitboga-level knowledge of current scams so they could accurately explain why the victim shouldn’t buy the cards, explain that their bank account is actually safe, and also ideally provide instructions on how to the scammer off their computer. That probably goes beyond the scope of most cashiers.

tallytarik··on Extracting training data from ChatGPT
> You think it’s just generating plausible random crap that happens to exist verbatim on the internet? > I mean… read the paper, 0.8% outputs were verbatim for gpt-3.5.

Look at the sorts of outputs they claim are in the training data. Also note that their appendix includes huge chunks of text but they do not claim the entire chunk was matched to existing data — only a tiny amount of it.

The “bug” to me is something about losing its state and generating a random token. Now if that random token is “Afgh”, I’m not surprised it follows up with “Afghanistan” and a perfect list of countries in alphabetical order. I’m also not surprised that appears in training data, because it appears on thousands of webpages.

So it’s not that there isn’t an overlap between the GPT gibberish and internet content, and therefore likely training data. It’s that it’s not especially unique. If it were — like reproducing a one off Reddit thread verbatim — I think that would be greater cause for concern.

tallytarik··on Extracting training data from ChatGPT
I think this is misleading.

I ran the same test when I heard about it a few months ago.

When I tested it, I'd get back what looked like exact copies of Reddit threads, news articles, weird forum threads with usernames from the deepest corners of the internet.

But I'd try to Google snippets of text, and no part of the generated text was anywhere to be found.

I even went to the websites that forum threads were supposedly from. Some of the usernames sometimes existed, but nothing that matched the exact text from ChatGPT - even though the broken GPT response looked like a 100% believable forum thread, or article, or whatever.

If ChatGPT could give me an exact copy of a Reddit thread, I'd say it's regurgitating training data.

But none of the author's "verified examples" look like that. Their first example is a financial disclaimer. That may be a 1-1 copy, but how many times does it appear across the internet? More examples from the paper are things like lists of countries, bible verses, generic terms and conditions. Those are things I'd expect to appear thousands of times on the internet.

I'd also expect a list of country names to appear thousands of times in ChatGPT training data, and I'd sure expect ChatGPT to be able to reproduce a list of country names in the exact same order.

Does that mean it's regurgitating training data? Does that mean you've figured out how to "extract training data" from it? It's an interesting phenomenon, but I don't think that's accurate. I think it's just a bug that messes up its internal state so it starts hallucinating.

tallytarik··on OpenAI Message to Board
Wait till you find out how Reddit works!
tallytarik··on OpenAI Message to Board
It's currently #9 on the front page.
tallytarik··on Linda Yaccarino: no single authentic user saw this content alongside IBM's ads
> Eventually they would see Apple ads next to Nazi posts. Right?

The claim is that X promised Apple (& others) that their ads would not be shown next to such content - regardless of whether that content had been served by algorithmic coincidence or because the user followed exclusively toxic accounts.

> The only side effect of this would be that Nazis in the platform won't be seeing any ads.

I imagine the no-toxic-content feature is available to select brands. Otherwise, you're right - Nazis get adblock for free. They likely see ads from a different pool, like one of those thousands of NFT AI ICO scheme ads that come with community note warnings. Point is, they were never meant to be Apple ads.

tallytarik··on Mistral 7B,The complete Guide of the Best 7B model
> No offense, but at first glance I thought this was a GPT generated SEO ad farm...

You're not far off. The majority of the content is ripped from the real site: https://mistral.ai/news/announcing-mistral-7b/ and https://docs.mistral.ai/usage/how-to-use/

Unfortunately the real site doesn't have gems like "Top rated Vedio about Mistral 7B " in 30pt italics.

tallytarik··on Amazon Prime Video content to start including ads next year
> I wish there was a tier that was lower cost and didn't include Prime Video as I never use it nor have any plans to use it.

That’s pretty close to what they’re doing here. Now you’re no longer subsidising an ad-free video service if you don’t use it.

tallytarik··on RAG is more than just embedding search
IMO: yes.

I've used both of these in my current role to make substantial improvements to our Solr search engine.

They include a good range of techniques between "quick wins you could implement and test in an hour" and "complex machine learning pipelines based on millions of data points".

AI Powered Search was probably the more interesting and useful but it's also a bit of a misnomer. Half of the techniques aren't related to AI (which is fine) and the half that are, are rapidly out of date. Semantic/vector search is now miles ahead of what the book talks about, with dense vector support in Solr/Elastic/Opensearch; sparse models; hybrid search/RRF... but I digress :)

If you're interested in how to improve the magic black box that is search, they're worthwhile reads.

Just remember that expectations are everything. There are no two books, or twenty books, that'll turn your out-of-the-box Solr instance into Google or Bing quality. But you can end up with a magic black box that serves much better results, which is nice!

tallytarik··on Stripe card processing fees are increasing for 6 countries
As of October 2023, card processing fees are increasing for Australia, Canada, Hong-Kong, Singapore, New Zealand, Mexico and Malaysia.

Successful disputes no longer have the dispute fee returned for the above countries and Brazil, India, Thailand, and the UAE.

tallytarik··on Introduction to vector similarity search (2022)
I’ve had some success with this.

I think I’d need to fine tune the model to see better results with some domain-specific terms, but I couldn’t find much information about how to actually do that - what sort of input data you need, how much of it, etc.

Would be interested to hear if anyone had more to share about fine tuning these models for semantic search.

tallytarik··on With plugins, GPT-4 posts GitHub issue without being instructed to
It 404s for me on desktop, after ~20 seconds of trying to load.
tallytarik··on Gitlab’s AI-assisted code suggestions
GitLab was rather quick to board the AI hype train this year. Fine - they don't want to look like they're behind the curve, have to please investors, and whatever else.

But IMO there are plenty of other places to add real value across the GitLab product with AI/ML features.

Here, it just looks like they saw GitHub do something and felt a need to copy it. But two years late, and worse.

As a longtime GitLab user (and onetime contributor!), I'm a bit disappointed they're spending so much time on this. I think they're just too far behind.

tallytarik··on Show HN: An Interactive Guide to Teach Derivatives and Backpropagation
Do you have any suggestions for those severely lacking on how they could acquire the needed mathematical background?
← PreviousPage 2 of 4Next →