I just tried a comparison of ChatGPT, Claude and Bard to write a python function I needed for work and ChatGPT (using GPT-4) whined and moaned about what a gargantuan task it was and then did the wrong thing. Claude and Bard gave me what I expected.
I actually considered starting a similar thread because I noticed this as well! Lately, it feels like GPT-4 is trying to get out of doing my work for me :D.
I've also started to notice it's been making a lot of typos eg. yesterday while converting Kelvin to Celsius, it incorrectly stated that 0C is -274.15K, despite correctly stating in the previous sentence the correct value of -273.15K.
Fraudulent activity by bots is a problem that everyone who's managed a large enough website is familiar with.
Most bots originate from cloud IPs or hide behind proxies and VPN services. Our scores ensure that the majority of legitimate users i.e. those coming from residential IPs are left alone, with captchas and other verification mechanisms shown to "high risk" users, that is, users visiting from a cloud provider's networks or a known proxy/vpn provider. This can be verified by the number of comments on this thread by users whose home internet IPs showed a high "Trust Score".
So if anything our scores help reduce the hassle or friction that legitimate users are subjected to by the fraud-fighting tactics employed by different websites.
IMO VPN services that don't defeat geolocation like Apple's Private Relay, the Google One VPN service and Cloudflare Warp are a good compromise for privacy.
This is because they allow businesses to provide their services without breaking the law eg. gambling is legal in some states and illegal in others, betting services need to distinguish/target users accordingly.
Insurance providers might only be licensed in certain states and not in others and also therefore need to correctly distinguish/target users correctly.
US companies are also restricted from doing business with certain embargoed and sanctioned countries and they are expected to use technological tools like IP geolocation to be compliant.
"OFAC makes clear its expectation that companies consider Internet Protocol ("IP") address geolocation data when assessing whether online customers are located in sanctioned jurisdictions." [0]
These and other legitimate usecases are defeated by VPNs.
This is actually proof that it works as intended. Our scores are made to be consumed by web applications.
In that context it makes sense that a cloud IP that's used to send mail would be treated with suspicion if it's seen trying to make a purchase on an e-commerce site.
For context, a few months ago we launched our Blocklists [0] feature which allows users to query 100+ blocklists and find all the ones where an IP address has been listed.
Blocklists based threat detection is however limited since they cannot contain every possible bad IP which leads to a lot of false negatives.
To fix this we created an IP Reputation scoring model [1] and currently provide 4 scores.
Time complexity doesn't seem like a great reason to use a bloom filter. Why not just check if the IP is in a set? Set membership is supposed to be O(1).
The reason to use a bloom filter would be to save space.
Read books. Find good books on the things you want to learn about and read them.
It might sound dumb but really all the stuff you want to learn has likely already been written up in beginner-friendly books. There are a lot of really good beginner/intermediate targeted books in Computer Science. You can get great books on Vue/React and any specific/niche technology as well as more broad books that cover wider areas eg. systems
Outside of school I've found it's a great way to deep-dive into a new topic and HN is a great place to get book recommendations for new topics.