Brug aldrig Google Gemini's Fast-model!

Google Gemini har tre modeller indbygget: Fast, Thinking og Pro (i den betalingsmodel jeg har - jeg ved ikke, hvad der er tilgængeligt i gratis-modellen).


Jeg fik en chok, da jeg så, hvor stor en forskel det gør. Først bad jeg den give mig et resume af en youtube-video (det kan Gemini som den eneste af topmodellerne) med Fast-modellen og bagefter med Thinking.


Det første resultat (med Fast) var komplet nonsens og havde intet med videoens indhold at gøre. Det andet resultat (med Thinking) var spot on.


Det er nærmest chokerende, så dårlig Fast-modellen er, og den bør aldrig anvendes til noget som helst efter min mening. Den er lige så dårlig som ChatGPT var, da den kom frem i november 2022. Jeg har før set folk brokke sig over, at Google ikke har fået udryddet hallucinationer ligesom de andre top-modeller, men dette resultat er værre end jeg havde forestillet mig.


Her er dokumentationen:


FAST: tldr: https://youtu.be/J9CWOH8I4Ow?si=fkOI7oxTGCP5-H5L


The video titled "The Rise of AI: How It Will Change the World" explores the transformative impact of Artificial Intelligence on society, the economy, and our daily lives. It balances the incredible potential of the technology with the ethical and practical challenges we face.


## Core Themes

  • The Intelligence Explosion: We are moving from "Narrow AI" (tasks like facial recognition) toward "General AI" (systems that can learn any task a human can).

  • Economic Shift: AI is expected to automate routine tasks, leading to significant productivity gains but also causing major disruptions in the labor market.

  • Human-AI Collaboration: The future isn't necessarily "human vs. machine," but rather humans using AI as a "bicycle for the mind" to solve complex problems like climate change and disease.


## Key Takeaways

Topic

Summary

Automation

AI will handle data-heavy and repetitive roles, forcing humans to pivot toward creative and empathetic skills.

Healthcare

AI's ability to analyze patterns will lead to personalized medicine and lightning-fast drug discovery.

Risks

Concerns include algorithmic bias, the spread of deepfakes/misinformation, and the long-term "alignment problem."

Education

Personalized AI tutors could democratize high-quality learning, tailoring lessons to an individual's pace.


## The Bottom Line

The video argues that AI is the most profound technology humanity has ever developed—surpassing the internet or electricity in scale. While it promises a world of abundance, the narrator emphasizes that the outcome depends entirely on how we choose to regulate and integrate these systems today.


THINKING: tldr: https://youtu.be/J9CWOH8I4Ow?si=fkOI7oxTGCP5-H5L


The video features Simon Sinek explaining a leadership lesson he learned from the Navy SEALs regarding the balance between Performance and Trust.

The Core Concept

Sinek illustrates this using a simple matrix with two axes:

  • Performance (The Y-axis): Technical skill and ability. Can you do the job well?

  • Trust (The X-axis): Character and reliability. Are you a good person?

The Navy SEALs' Perspective

When selecting elite teams, the Navy SEALs look at these categories:

  1. High Performance / High Trust: The ideal leader/team member.

  2. Low Performance / Low Trust: Obviously not a fit.

  3. Medium Performance / High Trust: Highly valued and often better leaders than high performers.

  4. High Performance / Low Trust: Described by SEALs as "The Toxic Leader."

Key Takeaways

  • Trust > Skill: The SEALs would rather have a medium performer they can trust with their lives than a high performer they can't trust.

  • The Business Flaw: Most companies have endless metrics to measure performance but almost none to measure trust. Consequently, businesses often promote toxic high-performers, which eventually destroys the team's culture.

  • The "Toxic" Test: Sinek notes that if you ask any team, "Who is the toxic person here?" they will all point to the same individual. Likewise, if you ask, "Who do you trust the most?" they will also all point to the same person—usually a consistent, reliable medium-performer who makes everyone else better.

The Bottom Line: Hire and promote based on character and trust, not just results. A high-performer with low trust is eventually a liability to the organization.


Kommentarer

Populære opslag fra denne blog

McKinsey As a Prompt

En tidligere OpenAI-medarbejder taler ud (positivt, men meget interessant)

Valg af den rigtige AI og den rigtige AI-model (Ethan Mulllicks seneste opslag)