Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

buyforlife.com

Post reply on HN
Print view View original

21–30 of 43 posts

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#21

Post by dragontamer »

bibabaloo wrote: post #16
woah wrote: post #14

One of the things that makes these GPT3 startups hard, and in my opinion most likely non viable, is that GPT3 is not ready for production. It can produce text that at a glance looks like it could have been written by a human, but a human that is not very competent at the task it set out to do and makes basic mistakes. For example: https://www.buyforlife.com/products/efa8b0c1-d7da-4485-b7f0-... - Blade will rust easil…

What am I missing? What is wrong with those? I read the complete review and honestly, if I didn't know better, would have thought that a human wrote it.

> - The locking mechanism is not as secure as it could be, and can be difficult to disengage when the blade is being used for heavy cutting tasks.

This sentence seems particularly bad.

A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use.

But... the next sentence says "hard to disengage during heavy use", which is the exact opposite (the knife will stay open after heavy use, and its difficult to close).

---------

- The blade is softer than most stainless steel blades, so the knife will dull quicker than others.

- Blade will rust easily unless oiled.

These two seem to contradict each other, but that's my knowledge of knife-steels talking. Softer steels are less prone to rusting but dull more quickly.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#22

Post by gjs278 »

bscphil wrote: post #15

Not really related to the GPT-3 thing, which seems cool, but a point of confusion and / or criticism: > if warranty = lifetime, then price/expected lifetime The #4 item showing for me is a tape measure, which costs $20 and has an estimated monthly cost of $.28. That means you are estimating this tape measure with a lifetime warranty will only last 6 years. The #3 item showing for me is an iron skillet, which costs $1…

in 10 years from now we will all agree the cast iron skillet enthusiasts were some of the most annoying. make sure you guys put it in the dishwasher to make it last awhile. avoid getting oil or salt on it, that rusts the iron. and definitely don’t scrub it, you’ll get iron filings in your food.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#23

Post by justusthane »

dragontamer wrote: post #21
bibabaloo wrote: post #16

Earlier quoted context omitted.

What am I missing? What is wrong with those? I read the complete review and honestly, if I didn't know better, would have thought that a human wrote it.

> - The locking mechanism is not as secure as it could be, and can be difficult to disengage when the blade is being used for heavy cutting tasks. This sentence seems particularly bad. A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use. But... the next sentence says "hard to disengage durin…

> A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use.

> But... the next sentence says "hard to disengage during heavy use", which is the exact opposite (the knife will stay open after heavy use, and its difficult to close).

I own Opinels, and these points are actually not mutually exclusive. The Opinel locking mechanism serves to keep the blade closed as well as keeping it open, and it is sometimes not as secure as it could be, causing the knife to open when you don't want it to (e.g. in your pocket), but it can also be difficult to disengage when the knife is open (still love them though!).

> These two seem to contradict each other, but that's my knowledge of knife-steels talking. Softer steels are less prone to rusting but dull more quickly.

I don't think this is necessarily GPTs fault—it seems to be a pretty divisive issue. If you Google now "stainless steel vs high carbon knifes", in the top few results you'll find articles claiming that stainless holds an edge longer and others claiming that high carbon holds an edge longer. I always thought that stainless held an edge longer while high carbon is easier to sharpen (and rusts easier, of course), but maybe I've been wrong on that.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#24

Post by mattrick »

dragontamer wrote: post #21
bibabaloo wrote: post #16

Earlier quoted context omitted.

What am I missing? What is wrong with those? I read the complete review and honestly, if I didn't know better, would have thought that a human wrote it.

> - The locking mechanism is not as secure as it could be, and can be difficult to disengage when the blade is being used for heavy cutting tasks. This sentence seems particularly bad. A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use. But... the next sentence says "hard to disengage durin…

Is this an issue with GPT-3 itself or with how it's aggregating multiple (potentially contradictory) reviews?

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#25

Post by dealforager »

Looks cool, but I do think there are ethical concerns with "scraping" reviews other people write and then not giving them credit. It reminds me of a talk I watched where a guy was saying how AI is essentially just taking work from others and taking the profits. For example, no one cares about the thousands of translators that lost their jobs due to AI translation, but those services were built on their work.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#26

Post by rasz »

>Summarizing product reviews

Dont know about you, but I dont give a flying F about random peoples reviews regurgitated by black box machine learning. When I buy something I either dont care about quality at all, or I look for an opinion of someone I actually trust.

Letting some company feed me recommendations is no better than believing in claims straight from product commercials. You are giving up your agency to a party treating you as a product.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#27

Post by nutanc »

Multiple things to think about here:

1. For summarizing, in my experience GPT-3 still has some ways to go. It gets it right a lot of the times, but when it misses, it misses bad.

2. Assuming that after scraping we feed all the scrapped data as a prompt from which GPT-3 generates bullet points, that will be a very big prompt. Since prompts are also counted as tokens, it might end up costing 10cents minimum to generate one summary.

3. I think the core USP from the process the OP has detailed is in steps 1,2 and 3. Step 4 is a good hook to get people to try this out, but have to test it properly and check the costs also.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#28

Post by picodguyo »

woah wrote: post #14

One of the things that makes these GPT3 startups hard, and in my opinion most likely non viable, is that GPT3 is not ready for production. It can produce text that at a glance looks like it could have been written by a human, but a human that is not very competent at the task it set out to do and makes basic mistakes. For example: https://www.buyforlife.com/products/efa8b0c1-d7da-4485-b7f0-... - Blade will rust easil…

I'd say the bigger challenge is building something that is defensible. It's not that hard to design a GPT-3 prompt that summarizes reviews or writes ad copy, so you quickly find yourself with a bunch of competition and not much to differentiate on besides price.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#29

Post by jcims »

justusthane wrote: post #23
dragontamer wrote: post #21

Earlier quoted context omitted.

> - The locking mechanism is not as secure as it could be, and can be difficult to disengage when the blade is being used for heavy cutting tasks. This sentence seems particularly bad. A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use. But... the next sentence says "hard to disengage durin…

> A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use. > But... the next sentence says "hard to disengage during heavy use", which is the exact opposite (the knife will stay open after heavy use, and its difficult to close). I own Opinels, and these points are actually not mutually exclusive…

Opinel also makes carbon steel blades that rust if not properly maintained. It looks like one of those reviews may have slipped in to the list.

Re: Show HN: Summarizing product reviews into simple bullet-point lists with GPT-3

#30

Post by nl »

dragontamer wrote: post #21
bibabaloo wrote: post #16

Earlier quoted context omitted.

What am I missing? What is wrong with those? I read the complete review and honestly, if I didn't know better, would have thought that a human wrote it.

> - The locking mechanism is not as secure as it could be, and can be difficult to disengage when the blade is being used for heavy cutting tasks. This sentence seems particularly bad. A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use. But... the next sentence says "hard to disengage durin…

> A locking mechanism on pocket knives holds the knife open. "Not as secure as it can be" implies that the knife closes in on itself (possibly snagging your fingers) under heavy use.

> But... the next sentence says "hard to disengage during heavy use", which is the exact opposite (the knife will stay open after heavy use, and its difficult to close).

I disagree entirely with this assessment.

There are plenty of badly made mechanisms that fail (eg, the knife flexes enough to slip past the lock) but are stiff and difficult to disengage.

As a specific example of a knife with this exact problem the CRKT M16-01 has complaints about it being stiff to disengage[1] and complaints about the lock failing[2]

[1] https://www.bladeforums.com/threads/frame-lock-way-too-stiff...

[2] https://www.bladeforums.com/threads/crkt-m16-14sf-lock-failu...

Post reply on HN