The idea that we are in danger of erroneously ascribing consciousness to AI sounds wild to me. Humans always assume whatever is convenient about other beings' (lack of) subjective experience as long as we have power over those beings. In this case the latter might no longer hold, but then it won't matter one bit what we think.
> How much would you pay to have access to a smarter version of yourself?
Call me obsolete, but I'd pay to interact with an intelligence that's as dissimilar to me as possible so I myself can grow from that interaction. The existing chatbots are already enough of sycophantic echo-chambers.
> all the vibe coded stuff is still slop (where’s all this new magical software that the productivity improvements should imply?)
Part-time vibe coder here. As far as I can tell it's no longer "slop" in the sense that I am no longer hitting the state where AI can't maintain what it has written or can't meet the requirements. But shipping is as hard as ever. Projects get more ambitious, scope creeps, nuances keep being discovered as you work on a piece of software, etc. Right now I feel that the absence of new software explosion is best explained by the fact that the part we have automated turned out to be relatively small.
Sorry, I understand scraping is a problem, but talking about open Internet while simultaneously complaining you can no longer discriminate datacenter IPs like you used to is hypocrisy.
I use a datacenter-based IPv6 address because my local ISPs don't offer v6 connectivity and the Internet is already broken for me. And generally the entire idea of a "residential" IP address smells.
Noone complained we can't discriminate DC IPs, though to be fair some (imo bad) operators did just that. This is not even about preventing bots, which has perfectly legitimate usecases (eg. Internet Archive).
This is about filtering out bad bots/actors who have no respect for your resources and will drain all of it causing bad experience for everyone. But because they know they don't respect robots.txt or even simple rate-limiting, they have to employ so-called residential VPNs. They're residential in that they route through real user connections, and so you can't block the IP/subnet without dropping a certain amount of legitimate human-driven traffic.
Personal example: some time ago, i had to disable a wordpress plugin on a site that was causing 100% CPU usage on the whole box (hosting dozens of wordpress instances). That plugin was a simple calendar, but a bot was repeatedly scraping non-existent (or rather, "no event planned for this day") pages for every date in the calendar that you can represent in the DB timestamp, clearing the cache as it went to try and find new events for 1000 years ago. Whoever operates this IP space doesn't matter to me, i'd just like to block them because they don't respect robot.txt… but i can't because they use a "residential proxy" and will change IP address every hour or so.
You could always try various counterattacks - returning a repeating compressed stream that decompresses to several terabytes, or returning one byte per second, or returning an endless list of hyperlinks to a honeypot, if the date is older than 200 years.
Sure, a lot of countermeasures is possible. I will definitely consider it next time, but for this calendar which had sitted unused for years on this instance, i decided to just turn off the plugin. A random wordpress plugin is never only a DoS avenue but also a bunch of potential vulns.
This is definitely an issue, where data centre addresses are, by default, second class citizens or even persona non grata. The problem is that this reputation has been, unfortunately, well-earned.
There's no easy answer here. The ephemerality and pseudonymity of VPS address usage screams untrustworthy, and the only way to reign that in is better identification of who is using the VPS/address or significantly more restrictive rules applied to data/port usage. And I'm not sure if I like the general direction that points towards - away from the "open internet".
The other possibility is that things get so mixed up that there's no way to tell identification at all and you have to allow everyone or no-one. That's the direction that residential proxies are pushing in.
This reminds me of the workflow I had a year ago. Miss Aider so much. Are there any good open source agents right now? Might be a good time to try one soon as Fable switches to token-based billing, which Code is designed to maximize.
Aider is still active. The new lighweight darling is pi. Complete batteries included solution, opencode. Turbocharge all of these and work on multiple harnesses at once, allowing every model to talk etc. Paseo. (Best mobile client too)
Around last fall it started chronically lagging in SOTA models support. Honestly, not sure why a code update is even needed for that. Eventually it'd get patched but it just wasn't quick enough for me. Other development seems to have ceased completely. At one point I tried to investigate what's going on and it looked like the community had moved on. There was a fork that promised a liberal approach to accepting changes and it looked broken beyond repair because of that.
It's disappearing. Even if models stop evolving tomorrow, there's still enough potential in the harness improvement to reach the point where anything 99% of us know about software engineering is useless. How many humans will be involved in creating software after the dust settles is anybody's guess, but I wouldn't bet on it being anywhere near the current level.
Sounds like calling those model "open source" did its wicked job. You can not take an open weight model and build a next-generation model using that as a foundation. Once those companies decide it's no longer in their interest to release new open weights everything you've created this way becomes a pile of rapidly deprecating legacy.
OP probably means using the existing open weights as a base for further homegrown development and research, not that the homegrown models are always updated based on whatever US or China are doing in the moment
you are not successfully riding the wave if you think prompting better is what they're talking about. They're talking about understanding the social structures your work happens in, how they're changing, and learning what you need to to be successful in that. Product and dev are getting closer together. Reviewing is becoming more important (and people don't have good theories about how to review, maybe you can make them.). New processes - and roles! are being created to deal with the vastly increased amounts of code produced, and the increased need for documentation and verification. Performance isn't a strong suit of the AIs, they get too rigidly stuck in existing architectures and fancy tricks, and often miss low hanging fruit that requires going and talking to some people to change the system. If you've got a statistics brain then metrics, logging and verification are booming (never met a manager who doesn't love a CHART). QA and testing are also bottlenecks of the new system, and unlikely to go away soon (ai can make tests, but it can't take the responsibility to say "this will work in production". You can!)
reply