Leipzig Gophers
blog

🔗Summer Drinkup #53 wrap-up

Misc

However, the influence of temperature on creativity is far more nuanced and weak than suggested by the “creativity parameter” claim; overall results suggest that the LLM generates slightly more novel outputs as temperatures get higher. – Is Temperature the Creativity Parameter of Large Language Models?

Randomness in embeddings: Do you get different embedding vectors from the same model on the same input? Yes and no. Quick test using ollama running on a Intel N3450 2016 6W TDP CPU (and 1.6W idle power consumption) reveals deterministic embeddings, whereas there are observations of the opposite as well:

Estimated cost of running this test (90 embeddings, total): at most 9W for about 6 minutes at 0.30 EUR/kWh: EUR 0.00027 (or framed differently: you can calculate about 640000 embeddings for the price of one scoop of ice cream - albeit it would take about a month to calculate them with this device) 🍦

If there’s a secret to this superhero-level productivity, it appears to have less to do with comic-book mutation and radioactivity, and far more with discipline, confidence, rigor, and many years of practice.

The positive effect of accountability was supported: those who sent weekly progress reports to their friend accomplished significantly more than those who had unwritten goals, wrote their goals, formulated action commitments or sent those action commitments to a friend. – The Impact of Commitment, Accountability, and Written Goals on Goal AchievementGoal Achievement


Join our meetup to get notified of upcoming events.