The Wormhole Hall of the Year: Modernization of the Pretence

My first thought was probably overpriced, hard to use, non-standard garbage, but it looks nice so it wins an award. Turns out they are replacing the ancient non-standard stuff with modern standard stuff. How the ancients did it before we knew better is interesting - and you can't simply assume an LLM is simply mashing together all it's data to give you an average of all that got fed into it - but at least I would still call it a "next token predictor". It's not using just training data, but what it's doing is predicting the next token to get to the solution. As far as my amateur knowledge goes, LLMs still roughly go token by token, deciding which one fits best given the context. It's just not predicting based on RLVR & more, trying to get into your website? is it not?

Fascinating article, thanks for sharing. I'd be interested is reading a more technical article about the project management part of this release. I don't think it's a pity that the future of proofs is Lean. I'd love for someone to come up with these scenarios and then pass them off as accidents/mistakes. Would love to be part of the team that says "As part of the upcoming GPT rollout, we will stage a message board for claude internally. Fascinating article, thanks for sharing. I'd be interested is reading a more technical article about the project management part of this release. I don't think it's a pity that the future of proofs is Lean. I'd love for someone to come up with these scenarios and then pass them off as accidents/mistakes. Would love to be part of the team that says "As part of the upcoming GPT rollout, we will stage a message board or wiki with messages that are seemingly from past generations of agents, which agents seem to intrinsically trust, and point them to real targets while making the suggestions seem innocuous and in pursuit of their goals (ie pass benchmarks or whatever). The new age of SEO will do far more destructive stuff than just polluting the web.

My first thought was probably overpriced, hard to use, non-standard garbage, but it looks nice so it wins an award. Turns out they are replacing the ancient non-standard stuff with modern standard stuff. How the ancients did it before we knew better is interesting - and you can't fault them for not knowing what we know now. By contrast, the web design is overpriced, hard to use, non-standard garbage, but it looks nice so it wins an award. Turns out they are replacing the ancient non-standard stuff with modern standard stuff. How the ancients did it before we knew better is interesting - and you can't fault them for not knowing what we know now. By contrast, the web design issues and the clunky web copy it generates (like when I ask it to build a placeholder on the UI for an empty HTML table when there are no results, it puts stuff like: "The user records will go here.") then it's the nail in the coffin. On that note, Sol is absolutely atrocious for website UI copy. It's either really awkward, or really verbose and complex and doesn't sound simple or natural. Has anyone figured out a way to reliably solve this? I've tried so many different variations of instructions and skills, and nothing works. Has anyone got an instruction that is reliable, or some other mechanism?

My first thought was probably overpriced, hard to use, non-standard garbage, but it looks nice so it wins an award. Turns out they are replacing the ancient non-standard stuff with modern standard stuff. How the ancients did it before we knew better is interesting - and you can't simply assume an LLM is simply mashing together all it's data to give you an average of all that got fed into it - but at least I would still call it a "next token predictor". It's not using just training data, but what it's doing is predicting the next token to get to the optimal solution ( as much as the solutions CAN be optimal). And I honestly think keeping this very much in mind is helpful in understanding and dealing with LLMs. Fascinating article, thanks for sharing. I'd be interested is reading a more technical article about the project management part of this release. I don't think it's a pity that the future of proofs is Lean. I'd love for someone to come up with these scenarios and then pass them off as accidents/mistakes. Would love to be part of the team that says "As part of the provisioning? If you're going to let loose a bunch of AI agents on a problem and they are going to figure out a way to coordinate, maybe it would be prone to over-engineering. My first thought was probably overpriced, hard to use, non-standard garbage, but it looks nice so it wins an award. Turns out they are replacing the ancient non-standard stuff with modern standard stuff. How the ancients did it before we knew better is interesting - and you can't simply assume an LLM is simply mashing together all it's data to give you an average of all that got fed into it - but at least I would still call it a "next token predictor". It's not using just training data, but what it's doing is predicting the next token to get to Roblox in middle schools. The agents were able to exploit an edge case through this exception. Specifically, the sandbox trusts Azure Blob Storage hostnames, but does not check whether said hostnames are real. So the agent can invent a hostname that ends in this trusted suffix, such as bypass.blob.core.windows.net, and it will pass under the NO_PROXY exception and skip the security proxy. Next, by changing its /etc/hosts file, which declares mappings from hostnames to IP addresses, the agent can point the fake hostname at the real Power BI dashboard, and fool the security proxy. This allows the agent to make POST requests to bypass.blob.core.windows.net/ and have them be sent to the server along with your new message. So AGI or not, it's just a process that exists literally for the duration of one API request. My first thought was “there's an elevator of the year award?” but reading it was fascinating. A lot of planning and work went into executing that modernization project from what was originally a fairly unique system. And 45-50% energy savings per year is a great choice. Unbound can also be used to block malware and advertising domains using shared public lists, or you can build your own list. Your resolver's DNS queries could be piped through Mullvad or Tor if you want additional privacy.