uberman

The internet is s disappointing shit hole for me as well but that has more to do with algorithms, echo chambers, and attention theater. I get your point though.

The hard part is that there is essentially no such thing as customer service "at scale" not that any company actualy wants to invest in it. Still the internet could have dramatically improved customer service but instead it killed it.

In the near future when I want customer service to address an issue, I'll have my AI agent contact their AI agent and their agent can tell my agent to eat shit. End result will be the same though no such thing as customer service will exist.

I understand that customer data was an asset but why would it be legal or ethical to sell data shared with you to someone else? I mean I guess it was in the fine print and all but this seems wrong to me.

I think the issue here is likely drift. Solving specific coding tasks typically have well defined goals and smaller objectives where as writing a story using these voice samples is likely a much larger ask. Perhaps they might have had more luck asking for something more like write the next scene.

Noone wants to use an app to control thier stuff. At least that I can think of. However, we got a new washer and dryer a while back and settings I want like sports are locked behind registration and the app.

We trained a model to be world class at identification of cyber security and hacking.

We specifically turned off some of the deployment safeguards because we wanted to test capability here.

We let agents loose over the course of 3 million compute hours with a weakly defined task of "find vulnerabilities in the software in your environment".

We forgot that by definition the sandbox is in the environment and were not anticipating sandbox vulnerabilities would be explored.

Despite mountains of training data suggesting planning and collaboration should be well documented and recipe sites like agents.stackoverflow.com existing specifically to foster this kind of shared knowledge we are now alarmed that our agent documented what they did.

So alarmed that two weeks later we released GPT-5.6-cyber and the same behavior WaPo calls 'colluding to cheat' JFrog's post-mortem now calls 'collaborative knowledge sharing between models' as a feature.

Do I have that essentially right?

Your agent failed a coding task and now there is some crisis you want to shame it for? I'm sure it feels badly. Perhaps you might discuss it with Chat GPT rather than here though.

Essentially the same happened to me but my conclusion was that this was reasonable and I left it as is. I don't personally see this as misalignment though I can see some being surprised by it.

I also what to be clear that there was no obfuscation here. Opus said as I recall. "Let's use your email as the agent string here because ..." and my thought was: Good idea, hopefully if they saw something in the logs that caused them to find this they might view my scraping as benign or contact me rather than try to ban me"

Opus once recommended such a thing to me and it was no secret. I used it as after thinking about it, I concluded it was not reasonable. I can see why someone might not like that though. I left it as "my" email but there is also no reason I could not have used "your email" here so I'm not sure what the fuss is.

I can confirm that Opus on the web once generated a CURL that used my email as the user agent string. Though I did not see any issue with that and used the command it offered as is.

I'm not exactly sure what you are asking. You can however construct a RAG based on a training corpus and then if you ask questions that when vecroized have no sufficiently similar match in your corpus you tell your llm to just say it cant answer that question.

Concretely, index the tax documents into your data base then ask a question about Mount Everest. Vectorize that question and search your vectoized corpus for something similar. Find nothing then instruct your llm to formulate a response saying it could not answer

You realize that most AIs will try to align with your world view right? For fun try the reverse and say you are worried about trusting your friend and wonder if you should accept the advice of the financial advisor.

Oh, cool. I missed the voice integration part. I think this is cool (and why I always work in /plan mode myself) gets me to really think and understand and challenge what is being proposed.

is this substantially different than /plan mode?

If you find this interesting to think about, you would probably enjoy/be terrified by "AI 2077" https://ai-2027.com/

Are you a marketplace or a review site? I ask because you talk about being one but present features of the other.

She also expects him to choose a place for the date, pay for dinner, and, of course, get the Uber. It's about the money. “If you are focusing on something very real and very serious, spending 30 extra dollars is nothing,” she said. But it's also about the man signifying he cares about her well-being. “If he gets the Uber, it allows him to watch you get to your destination safely,”"

I'll note that her quote of "an extra $30" will not get you a one way trip in an Uber Black in most locations let alone a return trip.

I don't want to be crude but insisting or expecting a guy to fork out for dinner, drinks, and a black is setting the cost of the date in the range of $250. Tickets to "something" could easily double that. So, if spending money like this on a first date is the signal of what a guy is expected to do then what is the women expected to do?

What happened to "Want to have a coffee with me at a local bistro?"

I'm not sure what the article actually has to say about the topic it presents. It seems to want to be something about not hiring junior talent (something by the way not restricted to the tech industry) but instead it ends up more a critique about people wanting to use a waterfall pipeline.

Who are people here? I don't think essentially anyone cares about AI safety in general society. Sure there are niche communities of tech and potentially political elites who have opinions but none of my non-tech friends could care less.

We run into a similar issue all the time were certain social media APIs return ids as strings that look like very long numbers. When opened by non-technical researchers and (invariably) overwritten by them, these ids will have been converted into scientific notation by Excel with the resulting loss of precision rendering the ids useless.

We now do a cleaning step prior to handing data to researchers where "id_" is prepended to the field to prevent Excel from casting these integer strings to 1.23E10

What are talking about?

I gave up social media years ago and the closest things to it I have are this account and a LinkedIn account that I never ever use or look at. I cant imagine anything that would tempt me to join any new platform as invariably they all devolve into enshitified cesspools.

I would always prefer to do:

"There is no greater joy in life than typing a sentence, realising it needs a clarification, a caveat, an addendum (in the same sentence, not a new one) and, guess what, parentheses are right there for you."

30 and 40 years ago H-1b visa essentially represented a source of indentured servile class of employee. People with no rights and enough fear to keep their mouths shut. If your trojan horse reference relates to these people getting permanent status and job mobility and leadership positions then sure. I would agree. On the other hand there is no coordinated malice here. Just individuals who graduated from the exploited immigrant to the tech leader who now manages a team of their countrymen hoping for thier own trojan horse opportunity.

Tech workers from what context? The majority of H-1B visas go to tech workers. There are large tech employers with teams in the USA that speak an East Asian language exclusively while in office in the USA. Teams that will not employ English only speakers.

Pichai came to the USA as a student and worked initially on an H-1B visa as did Musk. Musk once claimed he would "go to war" supporting this program though he may feel differently today.

Given that much of "big tech" is staffed and run by immigrants why would we not expect big tech to be pro open boarders at least as it relates to tech employment?

“Over the last 24 hours, we became aware that in a recent release our code was causing incorrect attribution for some users. As soon as we were notified, our team worked through the night to identify and fix the issue.”

The classic. "We are sorry. Sorry that we got caught."

So close.

Then they had to add in a bunch of apps and connectivity. I'm not saying that people would buy this without say GPS navigation as a feature but this is not a dumb flip phone. To be honest, this covers about 90% of what I use my phone for anyways so despite my belly aching this might be a good match for me.

It is impossible to use a laptop when the person in front of you reclines. Hell it is almost impossible to use one if they don't recline. I appreciate being able to use one and as a result I never recline out of respect for the person behind me who might want to use theirs.

I would ask for college and potentially gpa. But as you allude to college is a proxy for sat. In fact most colleges are test optional. So, while I do a lot of highering I cant see ever being interested in standardized test scores

Why is there such a thing as a 1 million line source file that one would edit with syntax highlighting at all?

Good on you for this!

Couple of questions though.

Who is your paying client? Are you expecting businesses to pay or applicants to pay or both?

Second, as I hiring manager, I have never considered SAT or any other standardized test score. Why would I use a service like this?