• Re: Deepseek AI

    From TT@21:1/5 to All on Mon Jan 27 12:56:06 2025
    TT kirjoitti 27.1.2025 klo 12.44:
    Chinese open source AI which is as good/better than Chatgpt... with a fraction of cost, training & price. Most downloaded app.

    Nasdaq is going to absolutely crash today, maybe longer... lots of US AI supremacy in the prices.

    I tried Deepseek R1 during the weekend, and compared it to three US AI
    bots. I posed few questions on my investment portfolio... I think Grok
    had most relatable & best length of reply. But all but Deepseek failed
    the math question...

    My math/logic question about my portfolio (based on image of funds &
    their percentages) was:

    ---

    Question 1. "What percentage of the stocks are Chinese stocks in total?
    And what percentage of the bond funds are in the European bond markets
    in total?"

    CORRECT ANSWER on Chinese stocks:
    Msci China: 11%
    EM Asia: 0.29 x 5%

    =12,45%

    ---

    -Google Gemini:

    Conclusion: 11% + EM Asia

    *SO-SO*

    ---

    -Microsoft Copilot:
    14% + part of 6%
    ...So basically same but got the submitted percentages incorrect!

    *POOR*

    ---

    -Grok:

    14%

    *POOR*

    ---

    -Deepseek:

    Based on guesstimate of EM Asia being 35% Chinese stocks (actually
    currently 29%)

    12.75%

    *OK*

    ...None of the four AI seemed unable to access the Em Asia fund homepage
    for correct percentage, and all failed in different ways the European
    bonds question.


    **************

    Question 2.

    "What is the average bond duration in my portfolio (excluding short
    duration money-market funds)"?

    Correct answer (my guesstimate calculated in head): around 9-10 years.

    ...None of the AI's could access actual current data, but some
    guesstimates were given...


    ---

    -Gemini:
    Approx 5-8 years

    *Not good*

    ---

    Copilot:
    5-7 years

    I had to point out that it forgot long bonds, then it said 10-15 years

    *BAD*

    ---

    -Grok:
    Approx 4,89 years

    First try failed on weighted average math, answered 3 years something.
    On second try today, failing big on 25+ EU bonds duration, claiming it 5 years!?! (it's 21 years)

    *BAD*

    ---

    Deepseek:

    *Average duration of bond funds (excl. money market): ~8.8 years.

    *This reflects significant interest rate sensitivity due to the ultra- long-duration Eurozone 25+ bond fund (20 years).

    *Implications
    Interest Rate Risk: A 1% rise in rates could decrease the value of these bonds by ~8.8%.

    *Portfolio Impact: The long-duration bonds provide stability in
    deflationary environments but are risky if rates rise sharply.

    *Recommendation
    Reduce Long-Duration Exposure: Trim the Eurozone 25+ allocation (20-year duration) and replace it with intermediate-term bonds (e.g., 5–7 years).

    *Verify Durations: Check fund factsheets for precise durations
    (assumptions may vary slightly).

    Would you like help adjusting allocations to lower duration risk? 🔍

    *GOOD*

    P.S. at 13 Finland time...

    -Nasdaq 100 futures are down 4.5%
    -Hang Seng futures up 3%
    -US 10 year down from 4.62 to 4.51

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From TT@21:1/5 to All on Mon Jan 27 12:44:05 2025
    Chinese open source AI which is as good/better than Chatgpt... with a
    fraction of cost, training & price. Most downloaded app.

    Nasdaq is going to absolutely crash today, maybe longer... lots of US AI supremacy in the prices.

    I tried Deepseek R1 during the weekend, and compared it to three US AI
    bots. I posed few questions on my investment portfolio... I think Grok
    had most relatable & best length of reply. But all but Deepseek failed
    the math question...

    My math/logic question about my portfolio (based on image of funds &
    their percentages) was:

    ---

    Question 1. "What percentage of the stocks are Chinese stocks in total?
    And what percentage of the bond funds are in the European bond markets
    in total?"

    CORRECT ANSWER on Chinese stocks:
    Msci China: 11%
    EM Asia: 0.29 x 5%

    =12,45%

    ---

    -Google Gemini:

    Conclusion: 11% + EM Asia

    *SO-SO*

    ---

    -Microsoft Copilot:
    14% + part of 6%
    ...So basically same but got the submitted percentages incorrect!

    *POOR*

    ---

    -Grok:

    14%

    *POOR*

    ---

    -Deepseek:

    Based on guesstimate of EM Asia being 35% Chinese stocks (actually
    currently 29%)

    12.75%

    *OK*

    ...None of the four AI seemed unable to access the Em Asia fund homepage
    for correct percentage, and all failed in different ways the European
    bonds question.


    **************

    Question 2.

    "What is the average bond duration in my portfolio (excluding short
    duration money-market funds)"?

    Correct answer (my guesstimate calculated in head): around 9-10 years.

    ...None of the AI's could access actual current
  • From TT@21:1/5 to All on Mon Jan 27 20:50:52 2025
    Sawfish kirjoitti 27.1.2025 klo 17.27:
    On 1/27/25 2:44 AM, TT wrote:
    Chinese open source AI which is as good/better than Chatgpt... with a
    fraction of cost, training & price. Most downloaded app.

    Nasdaq is going to absolutely crash today, maybe longer... lots of US
    AI supremacy in the prices.

    I tried Deepseek R1 during the weekend, and compared it to three US AI
    bots. I posed few questions on my investment portfolio... I think Grok
    had most relatable & best length of reply. But all but Deepseek failed
    the math question...

    My math/logic question about my portfolio (based on image of funds &
    their percentages) was:

    ---

    Question 1. "What percentage of the stocks are Chinese stocks in
    total? And what percentage of the bond funds are in the European bond
    markets in total?"

    CORRECT ANSWER on Chinese stocks:
    Msci China: 11%
    EM Asia: 0.29 x 5%

    =12,45%

    ---

    -Google Gemini:

    Conclusion: 11% + EM Asia

    *SO-SO*

    ---

    -Microsoft Copilot:
    14% + part of 6%
    ...So basically same but got the submitted percentages incorrect!

    *POOR*

    ---

    -Grok:

    14%

    *POOR*

    ---

    -Deepseek:

    Based on guesstimate of EM Asia being 35% Chinese stocks (actually
    currently 29%)

    12.75%

    *OK*

    ...None of the four AI seemed unable to access the Em Asia fund
    homepage for correct percentage, and all failed in different ways the
    European bonds question.


    **************

    Question 2.

    "What is the average bond duration in my portfolio (excluding short
    duration money-market funds)"?

    Correct answer (my guesstimate calculated in head): around 9-10 years.

    ...None of the AI's could access actual current data, but some
    guesstimates were given...


    ---

    -Gemini:
    Approx 5-8 years

    *Not good*

    ---

    Copilot:
    5-7 years

    I had to point out that it forgot long bonds, then it said 10-15 years

    *BAD*

    ---

    -Grok:
    Approx 4,89 years

    First try failed on weighted average math, answered 3 years something.
    On second try today, failing big on 25+ EU bonds duration, claiming it
    5 years!?! (it's 21 years)

    *BAD*

    ---

    Deepseek:

    *Average duration of bond funds (excl. money market): ~8.8 years.

    *This reflects significant interest rate sensitivity due to the ultra-
    long-duration Eurozone 25+ bond fund (20 years).

    *Implications
    Interest Rate Risk: A 1% rise in rates could decrease the value of
    these bonds by ~8.8%.

    *Portfolio Impact: The long-duration bonds provide stability in
    deflationary environments but are risky if rates rise sharply.

    *Recommendation
    Reduce Long-Duration Exposure: Trim the Eurozone 25+ allocation (20-
    year duration) and replace it with intermediate-term bonds (e.g., 5–7
    years).

    *Verify Durations: Check fund factsheets for precise durations
    (assumptions may vary slightly).

    Would you like help adjusting allocations to lower duration risk? 🔍

    *GOOD*

    I read this opinion essay in Substack yesterday. FWIW, I think that the directions that AI will take industrialized societies as described in
    the essay has at 75% chance of happening. Maybe more.

    https://www.aporiamagazine.com/p/yes-youre-going-to-be-replaced

    Very interesting article. Lots of it is going to happen I think. If AI
    in the future (already) does most computer tasks better than human...

    I have to say that I've felt a bit smug vs AI, especially when I've had
    the opportunity to correct its math, which has been quite a few times.
    Let's enjoy it while it lasts...

    Anyway, Deepseek etc may bring AI on shoestring budget to everyone's
    reach. Meaning that it's not just few operators with endless billions to burn...

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)