I just mean that the image input probably didn't affect the output at all. Could have just told it "I'm an amateur doing bulbs and batteries and it doesn't work" and it would give the #1 search result on forums for that which is "did you strip your wire."
I feel like I'm in the audience at a magician show, except most of the audience is breathlessly amazed and doesn't understand how easily tricked they are.
This really underscores how so much LLM "intelligence" is based off of people's experiences that they wrote about. It saddens me somewhat to see that it has basically all been captured by corporations now, and perhaps in a few years there will be little point in knowing things yourself simply because the LLMs will have gotten so good that there is no point to wasting the extra effort. Of course, this might lead to the atrophy of people's thinking muscles and a dumber and more subservient populace, but does anyone really care? Judging by the meteoric rise in the capabilities of LLMs over the past five years, is it really naïve to expect most knowledge work to be obsolete in the next ten?
The other stuff in this thread about decompiling firmware is a lot more interesting to me, though, seeing as it used to be a fairly demanding but rewarding task that has now been "solved" by Claude. It's a magic trick that is a lot harder to pull off than the other things in this thread.
This is getting ridiculous. One particularly concerning thing I’m seeing is that pull requests on both the web UI and API aren’t reflecting all commits or branch changes consistently. It would be very easy to merge something without realizing you’re not actually reviewing the full diff.
Yeah, I've had several occasions recently (seemingly not related to any incidents on the status page) where I've had to wait 20 minutes to an hour to be able to open a PR, because Github didn't recognize my branch had any new commits compared to the base branch.
I don’t fully agree. Yes, AI can be seen as a cheaper outsourcing option, but there’s also a plausible future where companies lean more on outsourced engineers who are good at wielding AI effectively, to replace domestic mid-level roles. In other words, instead of nullifying outsourcing, AI might actually amplify it by raising the leverage of offshore talent.
Consider the kinds of jobs that are popular with outsourcing right now.
Jobs like customer/tech support aren't uniquely suited to outsourcing. (Quite the opposite; People rightfully complain about outsourced support being awful. Training outsourced workers on the fine details of your products/services & your own organisation, nevermind empowering them to do things is much harder)
They're jobs that companies can neglect. Terrible customer support will hurt your business, but it's not business-critical in the way that outsourced development breaking your ability to put out new features and fixes is.
AI is a perfect substitute for terrible outsourced support. LLMs aren't capable of handling genuinely complex problems that need to be handled with precision, nor can they be empowered to make configuration changes. (Consider: Prompt-injection leading to SIM hijacking and other such messes.)
But the LLM can tell meemaw to reset her dang router. If that's all you consider support to be (which is almost certainly the case if you outsource it), then you stand nothing to lose from using AI.
> But the LLM can tell meemaw to reset her dang router. If that's all you consider support to be (which is almost certainly the case if you outsource it), then you stand nothing to lose from using AI.
I worked in a call center before getting into tech when I was young. I don't have any hard statistics, but by far the majority of calls to support were basic questions or situations (like Meemaw's router) that could easily be solved with a chatbot. If not that, the requests that did require action on accounts could be handled by an LLM with some guardrails, if we can secure against prompt injection.
Companies can most likely eliminate a large chunk of customer service employees with an LLM and the customers would barely notice a difference.
Also consider the mental health crisis among outsourced content moderation staff that have to appraise all kinds of depravity on a daily basis. This got some heavy reporting a year or two ago, in particular from Facebook. These folks for all their suffering are probably being culled right now.
You could anticipate a shift to using AI tools to achieve whatever content moderation goals these large networks have, with humans only handling the uncertain cases.
I see it the other way around. An internal person with real domain knowledge can use AI far more effectively than an outsourced team. Domain knowledge is what matters now, and companies don’t want to pay for outsiders to learn it on their dime. AI let's the internal team be small enough that it's a better idea to keep things in house.
In a vacuum, sure. But when you take two resources of similar ability and amplify their output, it makes those resources closer in cost per output, and in turn amplifies the risk factors for choosing the cheaper by cost resource. So locality, availability, communication, culture, etc, become more important.
Got it, Thank you. It's quite the hallucination, we'll continue looking into ways to keep the model on task! Please keep the feedback coming, these small models behave a bit differently from their large, cloud hosted counterparts.
I was thinking the same - high-quality content from across the web is increasingly paywalled, while low-quality free content flourishes. Only being able to afford to subscribe to a small number of sources keeps you inside the same echo chamber.
My mobile phone account has a bogus number (not generated by me!) as my SSN. Fortunately I know what it is so can recite it on a call. Which is why I don't want them to have my SSN in the first place!
It’s all fun and games until your bank reports something to the government under this SSN and it affects a person to whom its really assigned. Bonus points if “something“ has fiscal consequences for either of you.
Seriously, this is not kind of shenanigans you want to be doing for no reason at all, simply because the reason would be invented by someone else.
As my original comment says I don’t do this for anything tax-ish (for the same reason you mention). But there is no reason a phone company need know my SSN. Or the pharmacy.
That makes an unreasonable assumption that everybody across the whole industry is not just competent and can be trusted to have it together, but they also manage to achieve their important goals (which are not always having the cleanest code or architecture) using the same architectural dogmas as you do. One can define a datatype as FormattedValue | null, make a tuple of (value, present), have a special empty value or do all of that at once, because parts of the dataset were separately inherited during the merger.
I'm not saying it doesn't happen - I'm saying it's not the correct practice. When talking about financial services, this sort of garbage data could lead to compliance issues and financial penalties. Typically those mergers end up with remediation initiatives to at least ensure that all the invalid data is nulled out, or facilitate it's collection if it's critical.
As far as I am aware, that SSN is technically possible. Years ago SSNs could be traced back to the state/location of birth by looking at the first few digits (mine is like that), but they were changed to be completely random a while ago.
The problem with the US, and we're already seeing it play out with things like CCPA (California's GDPR-like privacy legislation), is that it is difficult to get momentum and bi-partisan support for sweeping changes in Congress. It either is watered down or gets deadlocked, so you end up in a situation where you have patchwork legislation, often conflicting in scope, state-by-state. I wouldn't expect much to change in the short to medium term.
Oh yeah but I mean the delivery guys are often underage themselves. It might present a legal issue. And indeed, they don't tend to be too precise, especially if you just hand them a nice tip :) Much less likely to happen in a physical store or bar.
So I could imagine the local law enforcement wouldn't have too much confidence in the process, that's what I meant :)
The main failure mode here I find is nuances in more specialized datasets. Elasticsearch is great, powerful and easy to work with once you get the hang of it, but is very open-ended with lots of room to build a sub-optimal solution.
Fuzzy matching and boosting often trip people up or lead to folks shooting themselves in the foot relevancy-wise.
If you want really great results, you need to spend the time crafting your query, dealing with synonyms ("pop" vs. "soda"), stemming, typos, negative boosts ("non-alcoholic", in the example) etc.
It's not necessarily hard, just often forgotten or not included in the initial scope.
There are numerous science fiction books that have explored the potential impact of artificial intelligence on humanity, ranging from utopian to dystopian scenarios. Here are some of the most interesting and relevant titles that address the questions and fears you've mentioned:
"Neuromancer" by William Gibson (1984): This cyberpunk classic envisions a future where AI is deeply ingrained in human society and plays a significant role in shaping it.
"2001: A Space Odyssey" by Arthur C. Clarke (1968): The novel, as well as the film adaptation by Stanley Kubrick, explores the relationship between humans and AI, particularly when AI begins to develop its own goals and intentions.
"Do Androids Dream of Electric Sheep?" by Philip K. Dick (1968): This novel, which inspired the movie "Blade Runner," questions the nature of humanity and consciousness as it relates to artificial beings.
"I, Robot" by Isaac Asimov (1950): A collection of short stories that feature the famous "Three Laws of Robotics" and explore various aspects of AI and their impact on society.
"The Moon Is a Harsh Mistress" by Robert A. Heinlein (1966): This novel features an AI that becomes self-aware and partners with humans in a fight for lunar independence.
"The Culture" series by Iain M. Banks: This series of novels is set in a post-scarcity society managed by benevolent, hyper-intelligent AIs called Minds.
"The Singularity is Near" by Ray Kurzweil (2005): Though not a novel, this non-fiction book predicts the future of AI, discussing the potential for AI to surpass human intelligence and the implications of this eventuality.
"Accelerando" by Charles Stross (2005): The novel explores a future in which AI, augmented humans, and post-humans coexist, addressing various ethical and societal issues along the way.
"Daemon" by Daniel Suarez (2006): This techno-thriller features an AI created by a deceased programmer that sets out to create a new world order.
"Altered Carbon" by Richard K. Morgan (2002): This cyberpunk novel touches on AI's impact on society, particularly in the context of human consciousness and digital immortality.
These books offer thought-provoking explorations of the possible consequences, both positive and negative, of advanced AI systems like ChatGPT. While each of these works is fictional and speculative in nature, they can still provide valuable insights and stimulate important discussions about the future of AI and humanity.
