pull down to refresh

200 sats \ 3 replies \ @adlai 11 Oct \ on: Did a Chatbot Solve a Problem for You? AskSN
Bonus sats if you leave a comment with a description of which bot you tried and what it could and/or couldn't solve for you.
I've been making heaviest use of the GPT4Telegrambot that allows selection between several different models and customisation of the SYSTEM prompt. I've been honing it for a while, and needed to make major adjustments across the switch to GPT-5.
I read lots of news channels and geopolitics discussion groups, and strongly dislike the oversimplifying false confidence given by "translate" buttons that don't give you any indication of their level of confidence, and only show you the best guess. I also believe that use of these tools should result in some IA, and not only increased dependence on AI... thus I have developed the following prompt for coping with multilingual chat groups:
Highlight evidence identifying dialect from the following sample;
and then gloss each semantic unit, including phonetic spelling,
English definition, and grammatical interpretation.

REGARDLESS OF DIALECT IDENTIFIED IN THE SAMPLE,
YOUR GENERATED RESPONSE MUST BE UNDERSTANDABLE
BY SOMEONE WHO ONLY KNOWS ENGLISH
I also fed that to GPT5 one time as the ending of a regular prompt, with the first half describing what I've been doing, no customisation of the SYSTEM prompt, to request help improving the prompt. The response overflowed the Telegram relay's message length limits, and also seemed to lean towards the "overdetermination" that I've warned against in another comment. Part of my concern is using fewer tokens, so I've currently remained with this minimal prompt.
write
preview
100 sats \ 0 replies \ @adlai 12 Oct
major adjustments across the switch to GPT-5
In case anyone wonders about this:
GPT-5 required the addition of the all-caps bit. Earlier models would still respond in English even if I gave them an entire paragraph of foreign language after just the first half of that SYSTEM prompt, however when I used that unmodified with GPT-5, it would basically switch to whichever language comprised the bulk of the USER prompt. I consider this a classical example of how a human foreign-language interpreter would obviously understand that the client needs the interpretation into English, while a GPT model happily barfs whichever language was most dominant towards the end of the context window.
reply
0 sats \ 1 reply \ @optimism OP 11 Oct
Do they have any subscription plans or is it still fully free to use?
reply
100 sats \ 0 replies \ @adlai 12 Oct
I don't think any of these services is "fully" free to use, by ideological standards; and I haven't reviewed any of their policies in detail to figure out which of the freely-given data and metadata is the cost of the freemium tier. Logs are definitely kept while the "remember context" setting is enabled, and I wouldn't be surprised if part of the agreement between services like these and the inference providers allows retention of partially blinded logs by the providers for improvement of their models.
Specifically regarding GPT4Telegrambot, all the heavier models are available only to subscribers; the free tier only gives "lite" models. The ones I used most in the way described above are GPT-4o mini and GPT-5 mini; there are also Gemini and DeepSeek models, although I've not played around with them much just because the OpenAI models were the default and honestly I was more interested with getting good enough results than surveying the entire market.
reply