PPlamenu
HomeTrendingLive feedsPeopleGroupsRulesStaff
Sign in
PPlamenu
HomeTrendingLive feedsPeopleGroupsRulesStaff
Sign in
Мя ��@mo@mastodon.ml
⁨17⁩d

if access to reasoning LLMs cost too much, or they run too slow on your hardware, this amazing lifehack can save you!

Here's how to use notepad.exe (or compatible) to execute your prompts:🧵

#AI #reasoning #LLM #lifehack

1000
Open original page
BoostsQuotesFavs
Мя ��@mo@mastodon.ml
⁨17⁩d

Replying to @⁨mo@mastodon.ml⁩

Start with an empty file. We will write markdown, as it's the most AI-native solution:

# How do I save money on tokens

Here goes your prompt as a header, making your goal clear, bold and visible

👇

1000
Open original page
BoostsQuotesFavs
Мя ��@mo@mastodon.ml
⁨17⁩d

Replying to @⁨mo@mastodon.ml⁩

Then you need to mark a start of reasoning section. Unordered lists are a great tool for that, each bullet point representing a short, but complete thought

- AI is billed by tokens
- Less tokens = less money to pay
- I need to use AI less
- And generate tokens myself!

👇

1000
Open original page
BoostsQuotesFavs
Мя ��@mo@mastodon.ml
⁨17⁩d

Replying to @⁨mo@mastodon.ml⁩

Keep imitating «chain of thoughts», until you get a reasonable basis to generate a complete answer to your initial prompt!

If you like and share this thread, in the next one I will teach you how to add memory and RAG to your notepad.exe-based inference engine!

⁨Jul⁩ ⁨29⁩, ⁨2026⁩, ⁨08:50⁩en
1000
Open original page
BoostsQuotesFavs
Мя ��@mo@mastodon.ml
⁨17⁩d

Replying to @⁨mo@mastodon.ml⁩

In case you are wondering, why this shitpost is a 4-toot thread, the reason is bluesky, and bridging to there

Their advanced decentralised network cannot handle posts larger than 300 symbols, because, uhm, decentralisation yes

0000
Open original page
BoostsQuotesFavs