Post New

https://www.perplexity.ai

Anyone else tried this site?

It's a large language model site that answers any question. So, search engine, not chatbot.

Why mention it? Unlike other models to date like what you usually see from ChatGPT, some differences:
* Information is up to date. They pull news feeds all the time.
* They pay the sources from whom they are pulling fresh data a cut from their own ad revenue. That's both nice, and a good way to have lots of partners on board with high-trust data. Early adopters include Time, Fortune, Der Spiegel etc.
* The results are relatively hallucination free: almost every sentence ends with one or more citation links so you can check the data source yourself.

I have found it surprisingly useful, so I just thought I'd pass it on.
And, oddly enough, clicking through to the citations has led me to some interesting sites I've saved for future use.

There is a free level and a Pro level, as you might expect.
It uses multiple LLMs. "GPT-4, Claude 3.5, Mistral Large, and others", some in Pro version only like Sonar Large 32k, reputedly known for its accuracy and conciseness. So, not like me.

I'm sure there are other projects making good progress on the accuracy problem, but this seems like a good step.
Everybody likes the answers from bots, except that bit about the answers being scarily wrong sometimes.

Jim

I like it. I asked several questions about geology and some obscure oil companies and got good answers. Then I asked about myself, and it knew nothing about my online activity, and it got my residence and phone number wrong. Perfect!

Pretty impressive. https://www.perplexity.ai/search/who-is-mungofitch...

I've been using it for 3 or 4 months after I heard a podcast guest rave about it. I like it a lot although, as one would expect, it's not infallible. Always good to double check its answers.

Then I asked about myself, and it knew nothing about my online activity, and it got my residence and phone number wrong. Perfect!

Work in progress, I guess!

And/or you have a doppelgänger.

Jim

Pretty impressive. https://www.perplexity.ai/search/who-is-mungofitch...

Yikes!
At least I didn't get doxxed. This time, anyway!
I prefer to be an International Man of Mystery. Baby.

Jim

It is much better than others I've tried.

But it's still very necessary to not trust, and to verify, everything an LLM so confidently says.
When I asked it to explain the 'five groves' approximation of intrinsic value of Berkshire, it initially provided 5 standard financial ratios as the answer. When I said that was wrong, it corrected itself and then basically got it right.

But it's the best I've seen, and I've bookmarked it, thanks Jim!

Agree entirely that all answers should be carefully checked before putting any weight on them.

With that said, I asked it about myself and got what might be described as a glowing obituary. Aw, gee.

But then I asked it a random question I'd never specifically thought of before:

"does any molecule naturally found in humans contain a benzene ring?"

(Brief pause)

"Yes, several molecules naturally found in the human body contain a benzene ring. Here are some key examples:

Amino Acids

Two essential amino acids found in humans contain a benzene ring: Phenylalanine: This amino acid features a benzene ring attached to an alanine side chain. It's crucial for the synthesis of other important molecules in the body. Tyrosine: Derived from phenylalanine, tyrosine has a benzene ring with an additional hydroxyl group. It's a precursor for several neurotransmitters and hormones.

Neurotransmitters

Several neurotransmitters in the human body incorporate a benzene ring: Dopamine: This important neurotransmitter contains a catechol group, which is essentially a benzene ring with two adjacent hydroxyl groups. Norepinephrine and Epinephrine: Both of these catecholamine neurotransmitters/hormones feature a benzene ring in their structure. Serotonin: This neurotransmitter, crucial for mood regulation, contains an indole group which includes a benzene ring fused to a pyrrole ring.

Other Biomolecules

Estrogen: The female sex hormone estrogen contains a benzene ring as part of its steroid structure. Thyroid Hormones: Both thyroxine (T4) and triiodothyronine (T3) contain benzene rings in their molecular structure.

It's important to note that while these molecules contain benzene rings, they are not the same as the toxic compound benzene itself. The benzene ring is a fundamental structural unit in organic chemistry, and its presence in a molecule doesn't necessarily imply the same properties or toxicity as pure benzene. These naturally occurring molecules play crucial roles in various physiological processes in the human body.


Accurate, as far as I recall (I would especially need to cross-check the neurotransmitter structures, but the rest are correct iirc.)

But it's as if you took put three or four top medical students in a room for ten minutes, and even then I'm not sure they would come up with all of these.

Just wow. Head and shoulders above the multiple sites of widely-varying quality that DuckDuckGo / Google would serve up.

(And the footnotes!)

Thanks again, Jim.

- sutton

I don't know if anyone else does this, but when I paste the title of a firewalled article to find it on another legitimate site it was shared with, if it's available I get at least one hit in the top 5 on DuckDuckGo/Google but not with perplexity.ai.
Probably pretty idiosyncratic....
It's killer otherwise.
:-)Shawn

It is much better than others I've tried.
But it's still very necessary to not trust, and to verify, everything an LLM so confidently says.

For sure. What I like about this one is the citations...it makes it easy to see where it's getting its information, so it really speeds up the validation.

I tried it on a programming problem. I asked for the C structure corresponding to a certain packet type of a communication protocol which is fairly widely used but not really publicly documented. It missed one one-byte field, admittedly one which is not generally used and I think is there mainly for alignment padding. The rest of the answer was correct and comprehensive, with good field names and descriptions. When I asked about the missing field, it apologised (!) and gave me the name and purpose of the missing field. So I couldn't use its response as given, but I'd call it a high quality fail. I added some of its response to the comments in a program I'm working on...but not the code.

Jim

I asked an electrical question where I knew the answer and it got it right and the explanation was good too.

Asked a second question about the intrinsic value of BRK stock and it had a number of answers, none of them obviously wrong.

I've been using Perplexity for a few months after a tip from of all people Morgan Housel, both the website and the app on my phone. I find using it with voice dictation saves time and typing. As others have said check the citations but don't assume that those original sources are right either. A statement on someone's website means nothing in itself and is significant only as part of a coherent pattern. Even Wikipedia is on occasion wrong.

But I find I converge faster to the truth with Perplexity than without it. It does what it's supposed to do; increase productivity.

I recently started checking out Perplexity after Jim suggested it, as I'd not heard of it before.

When I shared a bit about my present circumstance and described the lifelong interests fashioning my present identity, it offered to present a fictional story based on my contemporary experience, featuring recollections of past experience. I gave it the go-ahead.

It produced a tale that perfectly depicted my geographic environs, wrote a far-fetched story around what's going on now, and conjured rich memories of what I'd describe as a loosely plausible imaginary history, with nothing very specific beyond references to the very few characteristics I'd provided. While far from accurate, quite impressively construed.

I'm a long-time serious Excel guy. Today I asked Perplexity about an obscure MS Excel chart formatting problem I'd been perplexed with for weeks. It gave me the answer in a second by linking to a remedial Microsoft support exchange. I'm left grumbling at Microsoft and bubbling with Perplexity.

UnQ

It is very good for certain things.
I never go to a business website anymore for a real customer service number that the business really want to give out (until you have tried their answers page or chat bot) and therefore is hidden deep in their web pages. Perplexity answers that question immediately.
Recently, there was a question of causative fraud involving the sale of our corporate jet - I put in the aircraft information and was immediately told who the new owner was/is.
Whenever I buy something on line and the site asks if I have a discount code, Perplexity has been very good at finding ones that still work.
As a physician, it answers some very specific medical questions very well and provides the citations to check; yet other, broader medical questions, are poorly answered.
As a cook, nothing beats looking at the pages of images on a Google search to find a recipe that looks like what I have in mind.

Smufty

Well, it failed the common-sense test today.

I have an older, standalone system in my shop that has the habit of resetting its internal clock very time there's a flicker in the power supply. I don't get around to resetting it very often.

This morning, I wanted to check on an event from yesterday afternoon. Soon I'm asking Perplexity Pro:

"The clock is incorrect, saying it is 11:14 pm on Jan 10 2005 when the correct time is 9:15 am on Aug 30 2024.

What time did the incorrect clock show on Aug 30 2024 at 2:36 PM?"

It thought for maybe ten seconds.

"To determine what time the incorrect clock showed on August 30, 2024, at 2:36 PM, we need to account for the time difference between the incorrect clock's initial setting and the correct time. The incorrect clock was set to 11:14 PM on January 10, 2005, while the correct time was 9:15 AM on August 30, 2024. The time difference between these two times is 172,115 hours
. To find what the incorrect clock showed at 2:36 PM on August 30, 2024, we add this time difference to the correct time:

Correct time on August 30, 2024, at 2:36 PM
Add 172,115 hours

This calculation results in the incorrect clock showing 1:36 AM on April 19, 2044." (bolding original)

So, Perplexity:

- interpreted the question correctly, and
- applied the right correction, but
- in the wrong direction (See: reciprocal bearing, where one sails confidently directly towards the reef)

A bright fifth grader could have done this, but she would also know to apply a common-sense filter to the proposed answer before reporting it out.

-- sutton
(who knows he could have looked up the correct function in, say, Excel, and had the answer in a few minutes.
...instead, he's electing the pragmatic solution <especially after the week this market's had>: https://www.amazon.com/gp/product/B073Q48Z95/ref=o... )

Post New

Berkshire Hathaway FAQ

---

Contact Shrewd'm
Contact the developer of these message boards.