• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Merged Artificial Intelligence

If you ask it the one prompt first, will that influence the answer of the second prompt? (I would choose the accurate prompt first)
 
Generally a LLM will "learn" over the course of a session. If you start a new session it "forgets" what happened in the previous session.
 
A little more meat on the bones of Apple testing “reasoning” models. I’m uncomfortable with the use of the term “give up”.
pivot-to-ai.com

Apple: ‘Reasoning’ AIs fail hard if they actually have to think

Apple Machine Learning Research has a new preprint: “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity.” [Apple, PDF] “Lar…
pivot-to-ai.com pivot-to-ai.com
 
Ah back on my laptop and can now read the pdf. Interesting bit from the conclusions
Finally, we presented some surprising results on LRMs that lead to several open questions for future
work. Most notably, we observed their limitations in performing exact computation; for example,
when we provided the solution algorithm for the Tower of Hanoi to the models, their performance
on this puzzle did not improve. Moreover, investigating the first failure move of the models revealed
surprising behaviors. For instance, they could perform up to 100 correct moves in the Tower of
Hanoi but fail to provide more than 5 correct moves in the River Crossing puzzle. We believe our
results can pave the way for future investigations into the reasoning capabilities of these systems.
Click to expand...
I need to read up on LRMs.
 
Blue Mountain said:
First, there's the possibility of a "fencepost" error: If a child is born March 30 and dies on March 31, we can correctly say the child lived one day (31-30.) However, if the child was born and died a few hours later the same day, was the lifespan zero days or one? Some calculations include the day the child was born as a full day and others do not.

...snip...
Click to expand...
Copilot's "Think Deeper" response to: How many days did this child live, who was born on March 30, 1883 and died on September 15, 1905, given that 1888, 1892, 1896, 1899, and 1904 were leap years?

Let’s break this down step by step:

1. **From March 30, 1883, to March 30, 1905**, the child lived for **22 full years**.
- Standard years have **365 days**.
- Leap years have **366 days**.
- The leap years within this range were **1888, 1892, 1896, 1904**.

2. **Calculating Days in Full Years (1884–1904)**
- 18 regular years: \( 18 \times 365 = 6570 \)
- 4 leap years: \( 4 \times 366 = 1464 \)

3. **From March 30, 1905, to September 15, 1905**:
- March 30 to September 15 is **170 days**.

### Total Days:
\[
6570 + 1464 + 170 = 8204
\]

So, the child lived for **8,204 days**. Quite a precise lifespan to calculate!
Click to expand...

According to MS website Copilot think Deeper uses OpenAI’s o3-mini model, however if you ask the AI itself it won't tell you! There was no difference in Copilot between "Think Deeper" and "Quick Response"

I tried o3-high and it came back with this as its response:

First of all it came up with a function using python:

Code: 
import datetime,math,sysstart=datetime.date(1883,3,30)end=datetime.date(1905,9,15)print((end-start).days+1)

It then executed that code and came out with this output:

Code: 
8205

Then gave a text answer:
The child lived 8 205 days.

Explanation:

• Counting inclusively (both the birth day and the death day are included), the span from 30 Mar 1883 through 15 Sep 1905 is 8 205 days.

(You can verify the count manually year-by-year, or—more simply—use a date‐difference calculation as shown.)
Click to expand...

So that AI knew about your fencepost error.

Tried Claude Sonnet 4, that first told me its plan.

  • Calculate the number of complete years between March 30, 1883 and March 30, 1905, and identify which of these years were leap years
  • Calculate the number of days from March 30, 1883 to March 30, 1884 (first partial year)
  • Calculate the number of days from March 30, 1905 to September 15, 1905 (final partial year)
  • Sum up all the days: days from complete years (accounting for leap years) + days from first partial year + days from final partial year
Click to expand...

It used Python for its coding:

Python: 
from datetime import date# Birth and death datesbirth_date = date(1883, 3, 30)death_date = date(1905, 9, 15)# Calculate the differencetotal_days = (death_date - birth_date).daysprint(f"Birth date: {birth_date}")print(f"Death date: {death_date}")print(f"Total days lived: {total_days}")# Let's also verify by breaking it down year by yearprint("\nBreakdown by year:")# Given leap years in the periodleap_years = [1888, 1892, 1896, 1900, 1904]  # Note: 1900 is NOT a leap year (divisible by 100 but not 400)actual_leap_years = [1888, 1892, 1896, 1904]  # Corrected listcurrent_date = birth_dateyear_count = 0while current_date < death_date:    next_year_date = date(current_date.year + 1, current_date.month, current_date.day)        if next_year_date <= death_date:        # Full year        days_in_year = (next_year_date - current_date).days        is_leap = current_date.year in actual_leap_years        print(f"Year {current_date.year}: {days_in_year} days (leap year: {is_leap})")        current_date = next_year_date        year_count += 1    else:        # Partial final year        days_remaining = (death_date - current_date).days        print(f"Final partial year {current_date.year}: {days_remaining} days")        breakprint(f"\nTotal: {total_days} days")

It's Output was:

Code: 
Birth date: 1883-03-30
Death date: 1905-09-15
Total days lived: 8204

Breakdown by year:
Year 1883: 366 days (leap year: False)
Year 1884: 365 days (leap year: False)
Year 1885: 365 days (leap year: False)
Year 1886: 365 days (leap year: False)
Year 1887: 366 days (leap year: False)
Year 1888: 365 days (leap year: True)
Year 1889: 365 days (leap year: False)
Year 1890: 365 days (leap year: False)
Year 1891: 366 days (leap year: False)
Year 1892: 365 days (leap year: True)
Year 1893: 365 days (leap year: False)
Year 1894: 365 days (leap year: False)
Year 1895: 366 days (leap year: False)
Year 1896: 365 days (leap year: True)
Year 1897: 365 days (leap year: False)
Year 1898: 365 days (leap year: False)
Year 1899: 365 days (leap year: False)
Year 1900: 365 days (leap year: False)
Year 1901: 365 days (leap year: False)
Year 1902: 365 days (leap year: False)
Year 1903: 366 days (leap year: False)
Year 1904: 365 days (leap year: True)
Final partial year 1905: 169 days
Total: 8204 days
The child lived 8,204 days.

Born on March 30, 1883 and died on September 15, 1905, this spans 22 years and about 5.5 months. The calculation accounts for the leap years that occurred during this period (1888, 1892, 1896, and 1904).
Click to expand...
 
Last edited:
Wudang said:
Ah back on my laptop and can now read the pdf. Interesting bit from the conclusions

I need to read up on LRMs.
Click to expand...
Again their external behaviour fails the same way we see human external behaviour fails when presented with such puzzles.
 
Just seen someone post that Disney and Universal are suing Midjourney for copyright infringement. Good.
 
Disney and Universal will be looking for exclusive rights to drain their underpaid artists dry, all in order to train the AI things that will progressively make them even more underpaid. They are simply callous capitalist entities with no regard for artists.

Though I'm unsure if that's what theprestige was referring to.
 
Olmstead said:
Disney and Universal will be looking for exclusive rights to drain their underpaid artists dry, all in order to train the AI things that will progressively make them even more underpaid. They are simply callous capitalist entities with no regard for artists.

Though I'm unsure if that's what theprestige was referring to.
Click to expand...
If that were the only thing I was referring to, it would be a wash. I'm not going to get into the entire catalog of Disney's sins in this thread, though. Or at all, really. I authorize you to reach your own conclusions and form your own opinions.
 
Wudang said:
A little more meat on the bones of Apple testing “reasoning” models. I’m uncomfortable with the use of the term “give up”.
pivot-to-ai.com

Apple: ‘Reasoning’ AIs fail hard if they actually have to think

Apple Machine Learning Research has a new preprint: “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity.” [Apple, PDF] “Lar…
pivot-to-ai.com pivot-to-ai.com
Click to expand...
Thank you for that link. It helps explain many of the unrealised expectations and shortcomings of AI, especially LLM, that frustrates me when reading all the fluff from the AI facilitated communication.
The paper, “STOP ANTHROPOMORPHIZING INTERMEDIATE TOKENS AS REASONING/THINKING TRACES!” (PDF here)is particularly enlightening. But I’m probably biased towards its conclusions anyway.
 
theprestige said:
If that were the only thing I was referring to, it would be a wash. I'm not going to get into the entire catalog of Disney's sins in this thread, though. Or at all, really. I authorize you to reach your own conclusions and form your own opinions.
Click to expand...
New thread perhaps? I know that Disney is excessively litigious and jealous of its IP, but if there's an entire catalogue of sins, I'm curious.
 
You must log in or register to reply here.

Similar threads

Gord_in_Toronto
Artificial Intelligence Solves One of Archaeology’s Greatest Puzzles
Replies
16
Views
248
Explorer
Explorer

Back
Top Bottom