3 ways you NEED to use Fable-5

@remoteree

48.1K followers · usually 900 views a post

Views

5.7K

Likes

107

Comments

4

Shares

1

SoundTikTok sound · original sound - remoteree

Promoted product

Claude 3 Opus

Why it worked

The video leverages the novelty of a new, high-performing AI model and a limited-time promotional offer to drive interest. It also provides actionable tips for users, making the content valuable and shareable.

Summary

The video highlights the new Claude 3 Opus AI model, emphasizing its high performance in software engineering tasks and its lack of a temperature setting, suggesting it should be treated like a new hire. It also mentions a promotional offer for pro users and provides three tips for using the model effectively.

Structure

1Introduction to Fable 5 and its lack of temperature setting
2Comparison of Fable 5's performance and cost to other models
3Promotional offer for pro users
4Tip 1: The Ceiling Test
5Tip 2: Routing and avoiding trivial tasks

Original caption

A new tier of AI dropped yesterday — above the flagship, at double the price, scoring 80% on real engineering work. The detail nobody noticed: it doesn't accept a temperature setting. You don't tune this thing. You brief it — like a hire. **Have you re-tested what AI "can't do" this month? It's free until June 22.** #ai #claude #softwareengineer #techcareers #anthropic

More from @remoteree

Product placement

The video discusses and promotes Claude 3 Opus, an AI model, highlighting its capabilities and pricing compared to other models. It also mentions GPT-3.5 and GPT-4 Omni as alternatives for simpler tasks.

On-screen text

3 ways you NEED to use Fable-5 CLAUDE FABLE 5 DOES NOT HAVE A TEMPERATURE SETTING, AND THAT DETAIL IS THE ENTIRE STORY. FABLE 5 CAME OUT YESTERDAY AND I NEED YOU TO UNDERSTAND THIS. OPUS 4.9. IT'S A NEW DEER ABOVE OPUS. FABLE 5 SOLVES 80.3% OF REAL SOFTWARE ENGINEERING TASKS WITHOUT ANY HAND HOLDING. FOR REFERENCE, THAT NUMBER FOR GPT 3.5 IS 58%. FORGET BENCHMARKING FOR A SECOND. IT IS TWICE AS EXPENSIVE ON A PER TOKEN BASIS COMPARED TO OPUS 4.8. PER MILLION TOKENS THAT IT OUTPUTS, YOU $50. AND COMPANIES CARE BECAUSE THEY DON'T PAY FOR VIBES, THEY PAY FOR OUTPUT. BUT HERE'S THE GOOD NEWS USER THEN UP UNTIL JUNE 22ND, FABLE 5 IS INCLUDED WITHIN YOUR USAGE LIMITS. THAT IS INCREDIBLY GENEROUS, THINGS YOU CAN DO TO TAKE ADVANTAGE OF THE CEILING TEST. NO. 1 THE CEILING TEST. "Claude Fable 5 is the state of the art model on CursorBench. It's opened up a class of long-horizon problems that were out of reach for earlier models." Michael Truell CEO and Co-founder TAKE THE HARDESST TASK MODELS HAVE FAILED OR BUNGLED UP FOR YOU AND THROW FABLE 5 AT IT. NO. 2. YOU HAVE DO THIS. FIGURE OUT ROUTING. INTELLIGENCE COMES IN TIERS. YOU SIMPLY DON'T ASK A FABLE 5 CLASS OF MODEL TO DO TRIVIAL TASKS LIKE TAKE A BLOB OF TEXT AND CONVERT IT INTO JSON DATA. THAT KIND OF PEASANT WORK SHOULD BE RESERVED FOR MUCH CHEAPER MODELS LIKE GPT- 40 MINI OR SOMETHING LIKE THAT. AND NUMBER THREE THE YOU WILL GET AN API ERROR IF YOU MANUALLY PASS IN A TEMPERATURE PARAMETER TO FABLE 5. YOU DON'T GET TO TUNE THIS MODEL. YOU WOULD HAVE TO TRUST IT WITH YOUR WORK LIKE YOU WOULD A NEW HIRE.

Transcript

3 ways you NEED to use Fable-5. Claude Fable 5 does not have a temperature setting, and that detail is the entire story. Fable 5 came out yesterday and I need you to understand this. It is not like Opus 4.9. It's a new deer above Opus. Fable 5 solves 80.3% of real software engineering tasks without any hand holding. For reference, that number for GPT 3.5 is 58%. Forget benchmarking for a second. It is twice as expensive on a per token basis compared to Opus 4.8. Per million tokens that it outputs, you $50. And companies care because they don't pay for vibes, they pay for output. But here's the good news. For a pro user, then up until June 22nd, Fable 5 is included within your usage limits. That is incredibly generous. And here are three things you can do to take advantage of that window. No. 1, the ceiling test. Take the hardest task that previous AI models have failed or bungled up for you and throw Fable 5 at it. No. 2, you have to have to do this. Figure out routing. Intelligence comes in tiers. You simply don't ask a Fable 5 class of model to do trivial tasks like take a blob of text and convert it into JSON data. That kind of peasant work should be reserved for much cheaper models like GPT-4 Omni or something like that. And number three, the mindset shift. You will get an API error if you manually pass in a temperature parameter to Fable 5. You don't get to tune this model. You would have to trust it with your work like you would a new hire.