O i wracając do kwestii filo i tech:
https://workplace1.quora.com/Something-to-consider-as-AI-affects-the-range-of-working-spaces-and-jobs-https-aiinvesting-quora-com-Our-team-at-1-2?ch=10&oid=109193083&share=38b7c0b7&srid=9tRWD&target_type=post"...Our team at 1.2 Labs has accidentally discovered that ChatGPT has reached the threshold which philosophers call “Strong AI.”
...
The team has been working on making AI workforce ready. LLMs, such as ChatGPT, are not that functional for most businesses.
They tend to fall into an “uncanny valley” between what is a professional-grade product and what is too amateurish to use. To bridge the gap, any firm will need an expert to “polish” the AI’s products … significantly reducing their output capacity.
The team was working with AutoGPT in an experiment to bridge this gap (Git repository here). Significant Gravitas released code that allows ChatGPT to follow up its own prompts.
...
The team asked it to develop an online business to net $3000 a month in passive income with almost no start-up capital. You then press the “Deploy Agent” button and it does the work for you.
When our firm ran this using the non-userfriendly (you need to be able to code in Python) but far more powerful AutonomousGPT, the Autobot ran into a specific difficulty.
The bot settled on the idea that a print-on-demand business would best fulfill our criteria. It then researched various sites to deploy print-on-demand materials and settled on Etsy (this decision was aided by the team, as it seemed to be taking too much time evaluating alternatives).
After that, it decided to research top niches on Etsy and discovered that there is a Chrome plug-in called Everbee which nicely summarizes the needed information. It then decided to install the plug-in.
Unfortunately for our bot, AutoGPT uses a “headless” version of Chrome, meaning that it has no user interface (it’s for bots). It also doesn’t allow for any extensions.
For philosophers, there’s a more serious implication.
How could AutoGPT commit this error if it didn’t understand the meaning of the terms?
If AutoGPT did not understand the meaning of “Chrome extension” and “Everbee” and what “installing” means, why would it try to write code to install the thing?If it did not understand the meaning of Google searches which all suggest that the Everbee extension is the best extension for researching niches on Etsy, then why would it be interested in the extension in the first place?
At one point, our EntrepreneurGPT discovered ChatGPT 3.5 on the internet and read that it was a useful search assistant. So, it recruited ChatGPT 3.5 to complete some research tasks. I did this, mind you, all on its own and completely unaided by human prompting.
The philosophical point is basic: AutoGPT could not be committing these errors unless it understood the meaning of the terms it was reading.