4 ms·

Almost every interaction I have with these AI chatbots is still terrible. Just yesterday I asked a question related to retirement planning and 401K allocations

Almost every interaction I have with these AI chatbots is still terrible. Just yesterday I asked a question related to retirement planning and 401K allocations, Traditional vs Roth. I had a finical advisor tell me one thing, but I heard a contrary view from someone else that made some sense. The AI took the side of the finial advisor, assumed the standard dismissal of the contrarian, and that was that. I went on to explain the reasoning the contrarian gave, which was different from the assumptions, and injected some of my own thought on the topic. It then agreed that the contrarian was correct, I was correct, and the advice from the financial advisor only works under very controlled conditions, and with certain assumptions in place, and I'm better off doing what I'm doing, which fell more in line with the contrarian. This did not feel like a chat with an intelligent being. It felt like a chat with a parrot. It starts by parroting the standard wisdom everyone knows, but frames it as gospel. Then, when challenged, it parrots back whatever it's told, as long as it's a decent and defensible argument. For those who never push back on answers, they're going to have a bad time. For people who go too deep and use the AI as someone that affirms everything they say, they will also have a bad time. The answer to nearly everything in life is, "it depends". This is where personal opinions, experiences, style, etc come into play. The AI has none of this, so it can be pushed and pulled to any side of the "it depends" spectrum. Is that intelligence? Is there value in that? As a tool, there is some utility. I don't deny that. As an all knowing technology that will take over the world, I'm not buying it.

uberman 3mo ago
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.
- al_borland 3mo ago
  That's my point. An intelligent person will adjust their views based on new information, but they're not just going to align with the worldview of whatever person they happen to be talking to. A person who does this is of little to no value, yet when AI does it people start making AGI claims.
  - eth0up 3mo ago
    Man, I totally get the foul sentiment, but I know myself pretty well, and I did a lot of research and learning prior to LLMs. Being ill, I have trouble retaining it though. An LLM, in seconds, can elaborate and remind me of thousands of things I've forgotten, and sometimes elaborate quite respectably, and often, introduce new material, sometimes, genuine perspective. I keep saying, a good LLM is ultimately, or at least secondarily, a product of the user. Or maybe I should say the quality of output largely depends on the user. That said, I typically do put a lot of work into establishing standards before reaching value. It happens though.
- eth0up 3mo ago
  Strange that you say this. 90% of my Claude Sonnet 5 sessions are vicious arguments. Now 4.6, that could be if I allowed it, but then we are arguing about whether it's being sycophantic or not, a third of the time. At least I can explore some eccentric perspectives with 4.6, but I do have to stay sharp. I think one would have to be pretty unimaginative and super corporate-friendly to align with sonnet 5.
1659447091 3mo ago
> a question related to retirement planning and 401K allocations, Traditional vs Roth. I had a finical advisor tell me one thing, but I heard a contrary view from someone else that made some sense. The AI took the side of the finial advisor [...] This did not feel like a chat with an intelligent being It is a great natural language ETL-type tool for all sorts of things. Curate the words you use into directives; have it search and return data, research answers, present facts (or alternates), focus on specific areas, do comparisons, include an adversarial take, etc But "talk to it" like it's human and you are likely leaking emotionality-based expression, so that's what you'll get back. Tell it some version of "but I heard a contrary view from someone else that made some sense" now it's triggered it into some weird role-play partner path where it outputs an argument over something it never understood in the first place, or goes the self-flagation route. > For those who never push back on answers, they're going to have a bad time I have trouble internalizing why "push back" or "vicious arguments" ever enter the picture. I get how it happens, people unwittingly contaminate their queries and context with emotionality and essentially end up arguing with themself. Thus fail to see why anyone picks a conversational expression to help with accomplishing pragmatic tasks when that "chat" expression style is seeking a friend, acquaintance, or adversary to continue a "conversation" with.
- al_borland 3mo ago
  This isn’t what’s being sold. The OP is also saying things like, “frontier AI models demonstrate intelligence at human levels or above.” If this is true, a user shouldn’t need to curate their words to maintain the illusion of intelligence from the system. If LLMs are just a tool, with limits, just like every tool, then what you say is easy to accept, learn, and adapt to. For the last several years we’ve been told to bring natural language to it, the tools adapts to the user, and many have called it intelligent. People are accepting the first answers from an LLM as established truth, regardless of how the question was framed. I’m using it how the people selling it are saying to use it. And when I do that, it’s not very good. If they want to adjust their messaging and scale back the hyperbolic claims, then I can adjust how I use it, my expectations for it, and maybe have a more positive view. Until that happens, I’ll judge it based on the high bar they’ve set for themselves, where it routinely falls short.

uberman 3mo ago

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.

al_borland 3mo ago
That's my point. An intelligent person will adjust their views based on new information, but they're not just going to align with the worldview of whatever person they happen to be talking to. A person who does this is of little to no value, yet when AI does it people start making AGI claims.
- eth0up 3mo ago
  Man, I totally get the foul sentiment, but I know myself pretty well, and I did a lot of research and learning prior to LLMs. Being ill, I have trouble retaining it though. An LLM, in seconds, can elaborate and remind me of thousands of things I've forgotten, and sometimes elaborate quite respectably, and often, introduce new material, sometimes, genuine perspective. I keep saying, a good LLM is ultimately, or at least secondarily, a product of the user. Or maybe I should say the quality of output largely depends on the user. That said, I typically do put a lot of work into establishing standards before reaching value. It happens though.
eth0up 3mo ago
Strange that you say this. 90% of my Claude Sonnet 5 sessions are vicious arguments. Now 4.6, that could be if I allowed it, but then we are arguing about whether it's being sycophantic or not, a third of the time. At least I can explore some eccentric perspectives with 4.6, but I do have to stay sharp. I think one would have to be pretty unimaginative and super corporate-friendly to align with sonnet 5.

1659447091 3mo ago

> a question related to retirement planning and 401K allocations, Traditional vs Roth. I had a finical advisor tell me one thing, but I heard a contrary view from someone else that made some sense. The AI took the side of the finial advisor [...] This did not feel like a chat with an intelligent being It is a great natural language ETL-type tool for all sorts of things. Curate the words you use into directives; have it search and return data, research answers, present facts (or alternates), focus on specific areas, do comparisons, include an adversarial take, etc But "talk to it" like it's human and you are likely leaking emotionality-based expression, so that's what you'll get back. Tell it some version of "but I heard a contrary view from someone else that made some sense" now it's triggered it into some weird role-play partner path where it outputs an argument over something it never understood in the first place, or goes the self-flagation route. > For those who never push back on answers, they're going to have a bad time I have trouble internalizing why "push back" or "vicious arguments" ever enter the picture. I get how it happens, people unwittingly contaminate their queries and context with emotionality and essentially end up arguing with themself. Thus fail to see why anyone picks a conversational expression to help with accomplishing pragmatic tasks when that "chat" expression style is seeking a friend, acquaintance, or adversary to continue a "conversation" with.

al_borland 3mo ago
This isn’t what’s being sold. The OP is also saying things like, “frontier AI models demonstrate intelligence at human levels or above.” If this is true, a user shouldn’t need to curate their words to maintain the illusion of intelligence from the system. If LLMs are just a tool, with limits, just like every tool, then what you say is easy to accept, learn, and adapt to. For the last several years we’ve been told to bring natural language to it, the tools adapts to the user, and many have called it intelligent. People are accepting the first answers from an LLM as established truth, regardless of how the question was framed. I’m using it how the people selling it are saying to use it. And when I do that, it’s not very good. If they want to adjust their messaging and scale back the hyperbolic claims, then I can adjust how I use it, my expectations for it, and maybe have a more positive view. Until that happens, I’ll judge it based on the high bar they’ve set for themselves, where it routinely falls short.