Michael LarkinApr 18, 2024
Simon Willison
Love this theory by @alexhern that the reason ChatGPT uses words like "delve" a lot is that OpenAI outsource a lot of their RLHF annotation to workers in Nigeria, and Nigerian formal English has a slightly different vocabulary https://www.theguardian.com/technology/2024/apr/16/techscape-ai-gadgest-humane-ai-pin-chatgpt
TechScape: How cheap, outsourced labour in Africa is shaping AI English

Workers in Africa have been exploited first by being paid a pittance to help make chatbots, then by having their own words become AI-ese. Plus, new AI gadgets are coming for your smartphones

The Guardian
Apr 18, 2024 at 4:21pmWeb
Show threadRobert AtkinsApr 18, 2024
@simon @alexhern Fascinating. I’ll check to see if any of that RLHF work was done in Singapore and revert to you later.
Show threadDan Neuman 🇨🇦Apr 18, 2024
@simon @alexhern TIL that I talk like a Nigerian.
Show threadsabikApr 19, 2024
@dan613 @simon @alexhern
Also walk like an Egyptian?
Show threadLightning Bjornsson*  Apr 19, 2024
@sabik @dan613 @simon @alexhern I fear that soon, “bot” will become a racist slur
Show threadExcelAnalyticsApr 19, 2024

@simon @alexhern

So 'bot' is now a school yard insult. And "the generally flawless spelling and grammar are all what we’ll shortly come to think of as “robotic writing”."

Tough for us pedants.

Show threadKevin RiggleApr 19, 2024
@simon @alexhern @mekkaokereke so ChatGPT is learning English from ppl who are actually good at it, is what I’m taking from this
Show threadKevin RiggleApr 19, 2024
@simon @alexhern @mekkaokereke the only problematic part is that I really think we shouldn’t be devaluing the labor of people who can use and appreciate ‘delve’ in a sentence, which should be everyone
Show threadSimon WillisonApr 19, 2024
@kevinriggle @alexhern @mekkaokereke I enjoyed learning that 33 of the top 100 scrabble players in the world are from Nigeria
Show threadL'égrégore André ꕭꕬApr 19, 2024
@simon @alexhern I'm thinking it would be cool to see likely nationalities of authors in the graph.
My assumption is that the baseline / pre-chatGPT delvers are from non-UK/US (and the neo-delvers are global (e.g. include UK/US)).
Show threadTom Bellin Apr 19, 2024
@simon @alexhern "AI" is a mechanical turk, but instead of one short man in a large box, it's 20,000 Nigerians in a digital sweatshop.