Opus 4.8 is nearly unusable at the moment. Making extremely obvious errors, failing to debug issues even after being prompted with the exact line of code where the problem is. It should be illegal not to disclose serious degradations of performance while still charging full price. It may just be my region's datacentres degraded or something. The company is so opaque about what's happening, it's impossible to know what is actually going on.
The optimistic take is they're preparing to relaunch Fable 5 in the next day or so and siphoning off power for that. Perhaps they're doing a snap retraining to strip out any cybersecurity capability.
I've seen videos of LLM-based automated post generators running on Reddit and it's pretty horrifying. I've also witnessed large subreddits being entirely controlled by bad actors and users having no recourse. I'm definitely not leaving it solely up to human moderation, there are tons of automated detection mechanisms augmenting the human moderation and the system is being built upon every day. I know it won't be easy to solve, I'm keen that somebody should try to tackle this issue, because it's arguably the single biggest scourge of the internet right now and has serious real-world implications, like influencing the outcome of elections. The internet is real-life now, it should be tackled. The landscape is an arms race and there is too much benefit to be had by malign forces to stop trying to shape conversations with bot swarms.
As for residential IPs, I take your point that this complicates things even more. It is very difficult to know accurately which of those are being used as proxies. However, the site has a swiss-cheese model of defence, so if a user passes through one layer, they still come up against all the other layers, so losing one layer is undesirable but still manageable. Again, I cannot claim it to be perfect, but it's working decently so far.
I'm working on Topicle (https://topicle.com), a Reddit-inspired link sharing, voting and discussion platform that rethinks a bunch of basic aspects like moderation, anti-abuse and privacy to try to address some of the biggest pain points. Every moderation action has an admin appeal available, bots and astroturfing are aggressively monitored and policed, comment histories are always visible, no VPNs or datacenter IPs for write operations. Everything is self-hosted (Umami analytics) and there's no data leakage to big ad networks. The full feature set is listed here: https://topicle.com/about
There are also technologies new-ish to this kind of site included like every thread is a live thread via websockets, your post and comment scores update in realtime, notifications are realtime, you can DM other users and receive your messages immediately. So it's distinct from the everything is a hard page load world of 10 years ago and blurring into native software in a browser.
What I'm working on right now is a SwiftUI iOS app, because one of the most interesting observations from analytics has been that the internet is 70-80% mobile devices now, contrary to my 10 years out-of-date conception that people were mostly using the internet on desktops. So a mobile app seems non-negotiable to reach most users. I have a PWA already, but early users have repeatedly requested an official App/Play Store presence.
The stack is somewhat unique in that it's built with a Swift/Vapor framework (https://vapor.codes/) backend, with a more standard React Router 7 (SSR) frontend. I picked this framework mostly because I'm historically an iOS dev, but have found it to be very capable in its own right. I later discovered Apple themselves are using Vapor for some web services and have a team devoted to maintaining the server library (SwiftNIO https://github.com/apple/swift-nio) the framework is based on.
Anyway, it's very early still with launch via Reddit itself only 3 months ago. One of the biggest issues is getting it in front of people without appearing spammy and cold-start on a social platform is also brutal, you need users to get users, and round and round it goes. I may do a Show HN in the future if there's any interest in a real experience using Vapor as a production backend.
One of the major attack vectors is distillation, where millions of questions are auto-generated and coordinated to produce training data for new LLMs. Anthropic alleges Minimax, Deepseek and Kimi were trained this way. Deepseek 4 compares favorably to Opus, so they're probably trying to prevent Deepseek 5 from being a bootleg Mythos. https://www.anthropic.com/news/detecting-and-preventing-dist...
The leading hypothesis is the same one that explains why women get more autoimmune disease generally. Women mount stronger immune responses than men - protective in acute infection (men had worse acute COVID outcomes), but it comes at a cost: women are the large majority of lupus, MS, Hashimoto's and RA cases. If long covid is substantially autoimmune/inflammatory, as the autoantibody findings in the OP article suggest, the group already primed for autoimmunity is the one you'd expect to be hit hardest. Proposed drivers: immune-regulating genes on the X chromosome (e.g. TLR7) and estrogen being immunostimulatory where testosterone is suppressive.
>implying there is at least a substantial fake element to it.
The article actually argues against that reading: IgG transferred from patients into mice reproduced the symptoms. Mice don't have a nervous disposition. That points to a physical mechanism.
I caught this in the Dec 2023/Jan 2024 Covid wave, in a densely-packed Bay Area tech office. I only returned to near-full mental clarity in Jan 2026 - two years later. It's an insidious illness that needs more visibility. Poorly ventilated offices full of sick colleagues in close proximity are ideal conditions for transmitting airborne diseases, and it's far too easy to develop a debilitating chronic illness this way. There should be minimum clean-air standards for open offices to protect workers.
This is a distinction without a difference. Users were assured their selfies would not be retained and they were. Discord then proceeded to lose those selfies to bad actors, after promising not to retain them. The incident has caused enormous distrust of all age verification systems, which were already starting in the mind of the community from a base level of skepticism. It's already highly invasive to take a photo of yourself, but then the user must trust that the organization on the other end will handle it appropriately. To have that trust so conspicuously broken poisons the well for all other age verification systems and websites that are legally compelled to use it, or face penalties from aggressive organizations like OFCOM.
Not immediately deleting the selfie is a pretty fundamental and egregious mistake to make. People are particularly sensitive to selfies not being handled correctly after Discord lost thousands of them, despite promising to delete them after age verification occurred (and then not doing so) https://www.bbc.com/news/articles/c8jmzd972leo
The damage is limited because the selfie is only retained on device, but it still does not signal competency from the EU to fail at the most basic hurdle of disposing of the selfie once verification is complete.
People played Turtle because it was a superior experience to the paid official classic offering. It had properly balanced classes, tons of new, high-quality content, real support staff instead of bots with sub-5 minute wait time for service, policing bots properly instead of ignoring them. Blizzard could offer this quality of service but chooses not to.
Nobody is taking the side of the IRGC here, it's an awful regime that should fall in a just world. But it's inevitable they will retaliate against their neighbors, if their neighbors are complicit in attacking them. Qatar, Bahrain, Saudi Arabia, Kuwait are not innocent, they picked a side and are paying for it.
Its neighbors are hosting US bases which were used to launch attacks on Iran. Bahrain in particular hosted the largest US radar station in the region which was being used as the control centre to coordinate the attack on Iran [1]. These countries were absolutely not 'non-aggressors'.
Intel has just released a high VRAM card which allows you to have 128GB of VRAM for $4k. The prices are dropping rapidly. The local models aren't adapted to work on this setup yet, so performance is disappointing. But highly capable local models are becoming increasingly realistic. https://www.youtube.com/watch?v=RcIWhm16ouQ
All of Apple's software is rotting, from OSX to iOS to the developer tools. Massive bugs like the keyboard no longer working properly [1] are left unresolved for years. Whichever executive or engineering leader used to keep the quality bar high must have left or stopped paying attention. There also seems to be little culture of quality and ownership at the IC level.
They've just gotten some substantial amount of traffic of defectors from ChatGPT, so their system may be creaking under the load. They have previously said they don't degrade their models under excessive load, but the models sure behave like they do.
Selecting a server is one of the reasons the fediverse appears to have not seriously challenged incumbents. As soon as a non-technical user sees this, they bounce.
Other manufacturers are likely to use the same panel as the new XDR and Studio display. The peak display tech this year will be the same glossy, high refresh rate mini LED used in the Apple displays, sold by third parties for a more reasonable price. You have to compromise a bit on the design, but in return you get a sensible price, much greater input connectivity and 'dual mode' which is useful if you want to also use it for gaming.