Thursday, March 30, 2023

The IRS "Free Tax" scam and the hilarious reason why Turbo Tax is the only good free solution.

I wrote this for Facebook friends but I keep seeing people praising the "Free Tax" program and dissing Turbo Tax -- without any actual knowledge. So reprinting it here:

We did B's tax returns for 2022 using the IRS "Free Tax" program. This being America is kind of a scam. Sharing this as a guide to the unwary.

We used "TaxAct". She had state income tax forms for MN and CO. At the very end of the process you learn it costs $40 to print each state return. (Be careful when navigating, at first it seemed Federal eFile was not free; if you defer the state returns you just completed then the free option is available.)

MN is supposed to support Free File [1] but even after we removed Colorado from her form TaxAct still wanted $40 to print the state return. So reentering on the paper form.[2]

[1] Free File program are Arkansas, Arizona, Georgia, Idaho, Indiana, Iowa, Kentucky, Massachusetts, Michigan, Minnesota, Missouri, Mississippi, Montana, New York, North Carolina, North Dakota, Oregon, Rhode Island, South Carolina, Vermont, Virginia and West Virginia, plus the District of Columbia.

[2] If you start at the MN site they don't include TaxAct on their list of FreeFile options. So if one chooses a vendor from MN site you might get a better results.

Hilariously, if you hunt around you can find the free and effective Turbo Tax program for simple returns. It's free because Intuit is in litigation and need to keep it free until that's done

ChatGPT's 155 IQ and passing the Sebastian's children puzzle

A Scientific American contributor found ChatGPT (probably 3) scored 155 on a standard IQ test:

Estimated on the basis of five subtests, the Verbal IQ of the ChatGPT was 155, superior to 99.9 percent of the test takers who make up the American WAIS III standardization sample of 2,450 people. As the chatbot lacks the requisite eyes, ears and hands, it is not able to take WAIS’s nonverbal subtests. But the Verbal IQ and Full Scale IQ scales are highly correlated in the standardization sample, so ChatGPT appears to be very intelligent by any human standards.

We don't need to worry though because it couldn't answer this question:

ChatGPT easily fails at obvious riddles, such as “What is the first name of the father of Sebastian’s children?” (ChatGPT on March 21: I’m sorry, I cannot answer this question as I do not have enough context to identify which Sebastian you are referring to.) It seems that ChatGPT fails to reason logically and tries to rely on its vast database of “Sebastian” facts mentioned in online texts.

This is what happened when I tried the Sebastian test on GPT-4:

Ah, I understand now. This riddle is a classic play on words. The answer is "Sebastian" himself, as he is the father of his children.

Ok, so that's reassuring.

Update 4/12/2023: ChatGPT4 (C4) recently did well in some blinded advanced undergraduate exams.

A response to Scott Aaronson's rejection of an AI pause.

Scott Aaronson, who works on AI safety for OpenAI, wrote a critique of AI Pause that was not up to his usual standards. Here's what I wrote as a comment:

Hi Scott — I was confused by your post. I’m usually able to follow them. I won’t defend the letter directly and Yudkowsky/TIME is not worth a mention but maybe you could clarify some things…

1. 6m seems a reasonable compromise given the lifespan of humans, the timescales of human deliberation and the commercial and military pressure to accelerate AI development. Short enough to motivate urgent action, but long enough that reflection is possible. (I doubt we actually pause, but I agree with the principle. China isn’t going to pause of course.)

2. Let’s assume GPT 5 with an array of NLP powered extensions exceeds the reasoning abilities of 95% of humanity in a wide variety of knowledge domains. That’s a shock on the scale of developing fire, but it’s occurring in a hugely complex and interdependent world that seems always on the edge of self-destruction and actually has the capabilities to end itself. We’re not hunter gatherers playing with fire or Mesopotomians developing writing. There’s no precedent for the speed, impact and civilizational fragility we face now.

3. It’s not relevant that people who signed this letter were previously skeptical of the progress towards AI. I recall 10 years ago you were skeptical. For my part I’ve been worried for a long time, but assumed it was going to come in 2080 or so. 60 years early is a reason to pause and understand what has happened.

Lastly, I read the OpenAI statement. That seems consistent with a pause.

Tuesday, March 14, 2023

Everything is just fine.

The ad-funded internet is collapsing. It's been ailing for years but, as seems to the rule, the last bit is happening quite quickly. I'm guessing advertisers, who must be kind of dim, finally realized they were wasting their money. They did much better funding special interest magazines but those are gone now. Much of Silicon Valley has been funded by a now defunct business model. So maybe this is not a good time to be there. (Or, if you are young and have some money, maybe this is the best time to be there.)

Meanwhile human-cognition level AI has arrived. About 60 years earlier than I'd hoped. And some people thought "Don't Look Up" was about climate change. As a guide I recommend the movie Superintelligence; particularly the protagonist's approach to the end of the world. (Spoiler: It's a rom-com. Her world does not end.)

It's whitewater world all over now. On the bright side, we don't need to waste too much time on longterm planning. I retired in August of 2022 so I think I'll focus on mountain bike trail maintenance and, like Superintelligence's Carol Peters, on the needs of friends and family.

Sunday, March 12, 2023

I ask ChatGPT about the next step.

 Not a bad response.

”Describe what happens when ChatGPT is enhanced with natural language processing and domain-specific knowledge modules.”

--


When ChatGPT is enhanced with natural language processing and domain-specific knowledge modules, it gains several capabilities that allow it to better understand and respond to user input.


Natural language processing (NLP) is a field of artificial intelligence that focuses on enabling computers to understand and interpret human language. By incorporating NLP into ChatGPT, the model can better analyze and interpret user input, including understanding sentence structure, recognizing named entities, and determining the meaning of words in context. This allows ChatGPT to generate more accurate and relevant responses to user queries.


Domain-specific knowledge modules provide ChatGPT with specialized knowledge related to a particular field or topic. For example, if ChatGPT is enhanced with a medical knowledge module, it can better understand medical terminology and provide more accurate and relevant responses to medical-related queries. By incorporating domain-specific knowledge modules, ChatGPT can tailor its responses to specific user needs, making it a more effective conversational agent.


Together, these enhancements enable ChatGPT to better understand and respond to user input, providing more accurate and relevant information to users. This can lead to a more engaging and satisfying user experience, as well as improved productivity and efficiency in a wide range of applications.”