I was surprised by that part too. I always assumed that Craig was a young techy person.
My takeaway from this article:
- Craiglist was launched by a 40-year old, ex-IBM enginner. It started out as an online list for his circle of friends.
- The main competitor then was "Classifieds" section in newspapers. Craigslist out-competed them by simply being a better medium (No word-count limit + photo support).
- Meanwhile, newspaper executives failed to respond in a meaning manner. By 2010, 70% of their classified ad business was gone.
Craig seems to be fully retired by now, focusing on his philanthropy work, which I think is awesome.
This study finds 28% of U.S. workers use generative AI, saving users an average of 2.2 hours weekly and potentially increasing aggregate productivity by 1.1%. Usage is highest in computing fields, but these gains may not appear in official statistics since most adoption remains informal, with only 5.4% of firms formally implementing the technology.
I'm reading about the founding story of many startups .. at the early stage when they had few users, there was no moat. Success seems is a big moat. If this guy had taken venture financing(I know this wasn't a thing back then but humor me), trademarked milkbar, peppered ads everywhere, it isn't inconceivable that he'd get to 500 stores. Would that have led to lasting success? Unclear.
You really under-estimate how googleable 97% of customer service calls are. The average person does not make any attempt to solve their own problems before calling customer support. That's just life.
Yes in an ideal world we would have a live customer support representative for every function in every facet of society, but there are a limited number of human beings available for such things, and this is a pretty reasonable place to do a first triage using a LLM for very simple questions.
One of the most observed weaknesses of LLMs is that they have no clue when they're dealing with a difficult problem. There's no doubt that throwing an LLM at the problem would likely fix many simple issues. The question is whether or not it can accurately triage a difficult issue, which is a task they tend to struggle with.
When accuracy matters, answering a question incorrectly puts a person in an even worse situation than simply failing to answer the question.
ChatGPT is not trained to "escalate" an issue because there's nobody to escalate to. You can get this to happen pretty reliably via prompting, and with even light retraining basically 100%.
And here's the thing: most front-line customer service is also clueless about difficult problems. The IRS cannot pull 10,000 seasonal experts on the line, they are going to hire barely-trained part-time accountants who also flub hard questions.
But human brains have a more developed and reliable means of expressing uncertainty, which is still a challenge for LLMs.
e.g. part-time front-line customer service will prefix a statement with "uhhh..." if they don't actually know what they're talking about, even if they do have trouble answering accurately.
> e.g. part-time front-line customer service will prefix a statement with "uhhh..." if they don't actually know what they're talking about, even if they do have trouble answering accurately
You can literally prompt GPT4 "Prefix a statement with uhhhh if you don't know what you are talking about" and get similar behavior.
That doesn't mean the 'uhhh...' is related to the certainty of the remainder of the response.
I literally just tested your prompt, with the question "is the sky blue?" and chatgpt prefixed the response with "uhhh..."
These models create the illusion of thought by statistically stringing words together, but they don't actually think or perform judgement of their own.
Edit: After digging into this for a few minutes, I challenge you to try prompting an LLM to judge the certainty of its own responses. The results I am getting are even worse than I thought it would be.
So... 4o is not confident that only humans qualify as dependents?
I think even a very junior front-line customer service rep should be able to answer that one confidently.
It seems that what the model is actually doing is prefixing "Uhhhh" when your question is leading in a way that doesn't match the data it has. The fact that the IRS requires dependents to humans should be answerable with an extremely high confidence, and that data is without a doubt in their dataset... but again, the model doesn't actually experience human confidence or uncertainty.
Ultimately, the tax question you asked it is something simple for a front-line worker to answer. So either one of two things must be true:
* either GPT-4o is so bad at answering tax questions that it cannot even answer easy ones confidently
* or GPT-4o is so bad at determining its own confidence level that it doesn't know when it is able to definitively answer even an easy question.
Either situation makes it bad for this task.
As I mentioned above, humans are good for answering questions even when they don't know the answer, because they're good at expressing their confidence to other humans. In this case, you'd want the support agent to answer definitively that animals do not qualify as dependents. One could certainly make their chat bot answer unconfidently randomly, or in response to strange questions, or all the time, but then the confidence signal isn't actually providing social value of communicating certainty.
It's one of the reasons why I stopped joining facebook groups. Every day the same ^%$#^#%$ post by a [adjective] [derogatory term] who couldn't be bothered to use Google / Bing / ect.
It would have to be able to fix your records in the IRS database, not just give you advice from the FAQ like most LLM support bots. Which could be awesome, but it'd have to be robust against prompt injection attacks and other bamboozlement.
That's thinking inside the box of extremely low-effort propaganda.
The IRS is critical to stopping fraud and making sure everyone plays by the rules. Currently, we have folks not respecting the rules and getting more economic influence than they deserve. The current climate is an effective tax on honesty.
As long as budget increases to the IRS increase revenue to the government, they should keep growing.
If you take issue with individual taxes that you believe don't belong in our tax code, identify one and repeal it.
> As long as budget increases to the IRS increase revenue to the government, they should keep growing.
Why? The government is not a business. They should not be interested in 'raising revenue'. They should be interested in helping their constituents grow rich; not taking their wealth.
I think in this case, the IRS "raising revenue" up to the amount they "should" be getting based on the current tax law is a good idea. There is an upper limit.
This then makes me think that tax law may already account for "under collection", which might cause it to target "over collection". Increased IRS effectiveness can close this gap.
I agree that the government shouldn't need to be interested in 'raising revenue'.
It does however have to deal with malicious actors who ham-handedly try to game the system at everyone else's expense while patting themselves on the back for being clever. Man, if we could just fix this one thing, think of all the general operating costs that could be cut everywhere.
The US Government has been doing deficit spending and the national debt continues to grow at an alarming pace. To fix this, we can raise revenue and/or reduce spending. The easiest thing is to just improve enforcement of the existing tax laws (which are already agreed upon) via increasing the resources available to the IRS
reply