If you're intent on using agents, try having them work by themselves on one feature branch while you work normally on another. It feels like hiring a fast intern...
Highly opinionated: Regarding using agents for a main project or feature, as some do, I don't think there is such a thing as a "flow state." If you adhere to that working strategy, your hypothesis is that the agent is so much faster than you that you can get more done even without a flow state, even if the quality is a little worse, and your job has fundamentally changed from programmer to code reviewer. That will have its own set of skills for you to develop, such as efficiently reading their changes, managing costs and context, writing good prompts, etc.
Kagi Assistant's advantage of using their search engine, optionally fine-tuned to your preferences, instead of Google/Bing is so underrated, too. It's the killer feature for me
That position seems to have changed, I think? They are supposedly allowing people to buy Privacy Pass tokens without a Kagi account in-the-future™:
"Yes, this makes sense. This is possible because technically the extension does not care if you have an account or not. It just needs to be ‘loaded’ with valid tokens. And you can imagine a mechanism where you could also anonymously purchase them, eg. with monero, without ever creating an account at Kagi. Let us know here if you are excited about this, as it will help prioritize it."
https://blog.kagi.com/kagi-privacy-pass
They also provide a Tor hidden service which can be used with Privacy Pass (although you'd have to install their extension on Tor Browser for it to work)
Kagi is still a good solution for other use cases. For the privacy-oriented you have Privacy Pass, for the familymen you get very good parental control options, for college and gradschool you get the Academic Lens and the very underrated feature of "lensing" the AI assistant.
The Translator outperforms Google Translate and DeepL in my experience, too, and provides very nice context and such for translations. Kagi Maps might become the premiere OpenStreetMap interface, too, with amazing integration of other resources. Just one anecdote about it: I just checked in Kagi Maps a local restaurant I added a few weeks ago in OpenStreetMap, and on Kagi they somehow have some interior photos that circulated in a local magazine. Amazing stuff
I mean, are we supposed not to? This doesn't read like a blog at all, it even has the dreaded "Key Takeaways" end section... The content is good and seems genuinely researched, but the text looks "AI enhanced", that's all
So you attribute the golden era of Church influence, around the 12th century onwards, as the "not bad medieval era", yet the Church is somehow evil and not the new barbarian kings?
I mean, you should look citations up, because any self-respecting modern historian disagrees with most "dark ages" myths you were probably taught in school. We have even traced back the myths to their origins, lots of them being propaganda by French revolutionaries and puritans
Agree, either Vimium or Tridactyl or some other form of proprietary Vim bindings. Nothing feels quite as bad with browsers as programming with both hands on keyboard, poppping into the browser for a quick search and having to quick on stuff
To anyone working in these types of applications, are embeddings still worth it compared to agentic search for text? If I have a directory of text files, for example, is it better to save all of their embeddings in a VDB and use that, or are LLMs now good enough that I can just let them use ripgrep or something to search for themselves?
Funny quirk of the website: it says it needs JavaScript to work, but I can read everything and see everything just fine, and the formatting isn't even broken. Then I look at uBlock and 195 trackers are blocked. Yikes
If any other country should know better, you'd think it would be Germany. Ursula in particular married into a family which was both persecuted as Anabaptists and participated in persecution by being Holocaust enforcers, so how on Earth is she not aware of the slippery slope here?
If that passes it's over for the european internet, IMO. I would just shut off all personal online devices completely and go banking and work only on a cheap Xiaomi
This cope is officially dead now, not even Proton is believing Switzerland anymore. The pressure has gotten enough that they had to freeze all Swiss investment and start the process of moving key infrastructure to another country (I don't remember which, but it's the one Mullvad is at).
Truth is, Euros don't care about privacy. The endgame will probably be to host this stuff in the third world or something, like pirates do
Selfhosting Matrix might be a solution if this passes. The surveillance is to be installed at the app level, imposed on the distributing companies (say, the Signal front-end), this is not a ban on. But if you're booting up your own application, it might at the very least be a legal grey area whether or not you need to implement chat control, so you could just not and the data will still be E2EE in travel for now. Easier than asking everyone you know to use GPG
As I understand it, the pressure for a surveillance state comes from the EU, right? While Germany and countries in their cultural sphere like Austria are fundamentally opposed to that. If that's the case, it won't ever change unless the EU changes
Only in the small subset of programmers that post on HN is that the case. Most users or even most developers don't mind slow stuff or "getting into flow state" or anything like that, they just want a nice UI. I've seen professional data scientists using Github Desktop on Windows instead of just learning to type git commands for an easy 10x time save
I haven't seen it happen at all with RAG systems. I've built one too at work to search internal stuff, and it's pretty easy to make it spit out accurate references with hyperlinks
"thoughtful" is the key word that makes it not a contradiction logically speaking, as the author likely was writing about precisely the stress between meaningful participation leaving a legacy and mindless social media consooming.
Arguing for Claude, though, if you think of "contradictions" in the more wishy washy continental philosophy way, it fits. There is a stress between the concepts of "digital footprint" in the mass surveillance/commercial capitalism sense and "legacy" in the writer/creator sense, and it is a cool distinction to point out, where one ends and the other begins. If you read the full quote by Claude, it seems to be leading to this, especially the passage below:
"This reflects a deeper tension: How do you engage meaningfully with systems whose fundamental nature you find problematic?"
Maybe their excuse will be that it can't be proven whether or not the deep learning model is discriminating based on this or that, it's just taking in as much data as possible?
This seems like a good use case, I'm optimistic on this one. But it smells fishy how often OpenAI releases these secondary products like custom GPTs, tasks, etc. It's looking like they know they won't be an LLM provider, like the YC sphere hoped, but an AI services provider using LLMs
We can assume that OpenAI/Anthropic offerings are going to be better long term simply because they have more human capital, though, right? If it turns out that what really matters in the AI race is study mode, then OpenAI goes "ok let's pivot the hundreds of genius level, well-paid engineers to that issue. AND our engineers can use every tool we offer for free without limits, even experimental models". It's tough for the small AI startup to compete with that, the best hope is to be bought like Windsurf
I have the exact opposite idea. I want the tests to be mine and thoroughly understood, so I am the true arbiter and then I can let the LLM go ham on the code without fear. If the tests are AI made, then I get some anxiety letting agents mess with the rest of the codebase
In this case it seems more like incompetence mixed with classic Euro bureacracy. The suits don't know better and consumers are braindead so won't even notice