Everyday, I look at IPs and try to improve the proxy / VPN detection system. Small changes here and there happen often and I test various datasets. If I were to tweet them it would be too spammy. Rest assured the API is well maintained and constantly improving.
LLMs getting kinda crowded. Next major step will be nanoLMs, running on < 16 GB storage, 4GB RAM, 15k context window on the device. If possible without ASIC hardware. Let the optimization begin.
Updated the proxy / vpn detection web site and web interface to look more “modern". It used to look like the man page from 2000s but now looks like a man page from 2010s 😎
Been working on GetIPIntel for many years now, I'd say it's pretty much complete except some maintenance and R&D. I can now relax a bit and I'm thinking about delegating some tasks like billing so I don't have to do everything.
GetIPIntel's API & website did not have any downtime with the recent cloudflare outage. Internal & external monitoring showed nominal status.
Incident reference: https://t.co/PFt3si1sVm
A system that can be easily spoofed with proxy / VPN can be arguably worse than not publishing that information at all. It provides a false sense of reality based on authority, and that authority operates like a black box. In the end, it might achieve the opposite intended effect
I suspect that proxy / VPN usage is going to increase greatly on X / Twitter with this new additional geolocation account information. Interesting days coming for X's dev team. Hopefully they'll publish some insight on their confidence level of geolocation information.
Had some downtime today due to cloudflare issues. Backup systems were triggered for people with 2N fail-over. Raw logs suggest no downtime as the last line on production and first line on backup overlapped. 2N users might've experienced some intermittent issues.
Added a new feature: oflags=r
Include residential proxy detection data via https://t.co/Q51xuAejT7 to generate the overall score. Full documentation is available on the website.
It is an opt-in feature, if there's any issues / comments, let me know.
A monitoring website report that the API is not accessible but I was logged into the system the whole time working on some code. Automated backup system also shows no issues. Cloudflare monitoring shows no problem at the moment. If you have problems accessing the API let me know.
Multiple Cloudflare services, including Workers KV, Access, WARP and the Cloudflare dashboard, experienced an outage for up to 2 hours and 28 minutes earlier today. Here's a detailed breakdown of what happened: https://t.co/5vbVg2MKIl
Currently @cloudflare is experiencing some technical issues. The services I use appear to work still with perhaps a degraded performance, adding ~25 ms of latency. It could also be network congestion amongst other factors so nothing for me to do but monitor the situation.
Google's AlphaEvolve is using a technique that's basically applying natural selection in an infinite loop to get some incredible results. Given the achievements that are made so far, it is a step forward not just AI / AGI but super intelligence.
https://t.co/DYl75qT25O
GetIPIntel also uses a structured format and I've taken great steps to avoid data poisoning. One would think if 1/100 data is bad, the negative effect would be 1% but in reality it's ~10% degradation in confidence. These concepts were implemented in GetIPIntel way back in ~2017.
DeepSeek R1's optimizations are impressive and frankly necessary for AI advancement. The training process was meticulously planned like providing structure and avoid data poisoning. The juicy part is the reinforcement learning, which is essentially an optimization function.
DeepSeek R1's optimizations are impressive and frankly necessary for AI advancement. The training process was meticulously planned like providing structure and avoid data poisoning. The juicy part is the reinforcement learning, which is essentially an optimization function.
I know I haven't tweeted a lot recently but I'm still actively working on GetIPIntel. This year, I've tested hundreds of different datasets and only about 4 of them were incorporated. The backend now parses on average ~1 billion pieces of information per day.
@MagicMirrorAI It allows the algorithm to have good data to work with when generating the score. I'm always playing with data and testing to see what works best. It's a time consuming process but it only has to be done once and everyone can benefit from it.
@MagicMirrorAI I'm glad you like the service. The system uses machine learning & probability theory algorithms to generate the score in real time; it doesn't rely on blocklists. The backend now parses around a billion pieces of information per day on average.
I have some tier 2 infrastructure in the southern east coast so I'll be monitoring #HurricaneMilton. There are redundancies in place so no downtime is expected. Stay safe out there.