- AI Tool Report
- Posts
- š”ļø OpenAIās new safety approach
š”ļø OpenAIās new safety approach
š„Mistral launches AI model
TOGETHER WITH WEBAI
Welcome to AI Tool Report!
Thursdayās top story: OpenAI has revealed a new AI-based approach to ensuring its AI models adhere to safety protocols.
š¤ļø This Morning on AI Tool Report
š”ļø OpenAIās new safety approach
š How to see the future of AI with WebAI
š„ Mistral vs Meta: Introducing Large 2
š½ļø How to make social videos in minutes with Augie Studio
š How to adapt a brand strategy with ChatGPT
š Microsoft vs Google: AI search
Read Time: 5 minutes
STOCK MARKETS
![Stock tracker](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e71b0bba-c336-4700-a7ac-62ca5216c568/430.16__38___1_.png?t=1721890466)
ā¬ļø The tech sector and AI stocks suffered badly during Wednesdayās session. Telsa was the biggest loser of the day down 12.33%, NVIDIA lost 6.8%, and Microsoft 3.59%. This price correction could be the start of a sustained reversal. Learn more.
ā ā ā ā ā ā
SAFETY
š”ļø OpenAIās new safety approach
![OpenAIās new safety approach](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/b473156c-9f51-47c0-a620-974c8586eed6/OpenAI_s_new_safety_approach.jpg?t=1721897635)
Our Report: OpenAI has developed a new way to improve the safety of its AI modelsācalled Rule-Based Rewards (RBRs)āthat uses AI to align model behavior with specific safety standards and policies, without human intervention.
š Key Points:
Previously, humans would score the AI model's responses to prompts based on how accurate they were or which they preferred: a method that was costly, time-consuming, and vulnerable to subjectivity.
With RBRs, safety teams can create rules for the model and AI will score its responses based on how closely they align with these rules, which is more efficient and non-subjective.
During testing, RBR-trained AI models showed improved adherence to safety standards and reduced instances of incorrectly refusing to answer a prompt, compared to those trained using human-led feedback.
š¤ Why you should care: While RBRs are a step forward in making sure AI models remain aligned with desired safety protocolsātherefore creating safer modelsāOpenAI has acknowledged that while RBRs could reduce training time, cost, human oversight, and subjectivity, using AI to guide its models could potentially increase bias, so safety teams must design RBRs carefully āto ensure fairness and accuracyā and consider using them in conjunction with the traditional human-based feedback approach.
ā ā ā ā ā ā
šļø WebAIās biggest release yet
Get an insiderās look at the future of AI at WebAI's Summer Release event
Discover how businesses are using WebAI to:
Deploy some of the world's largest models locally on distributed networks
Train and customize models quickly and easily
Uphold data privacy and IP protection by building and deploying locally
TRENDING TOOLS
Meco is a distraction-free space for reading and discovering newsletters, separate from the inbox
DreamFace uses AI to enhance and animate facial features in photos
Bigjpg enlarges and enhances images without compromising quality
Anime is an AI-powered tool that generates anime art
Ebsynth is an AI video editing tool
ā ā ā ā ā ā
LAUNCHES
![Mistral vs Meta: Introducing Large 2](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/5e7e005d-ad09-44f3-9f27-d9283a46183b/Mistral_vs_Meta_Introducing_Large_2.jpg?t=1721897856)
Our Report: Just one day after Meta revealed its newest (and biggest) AI modelāLlama 3.1āFrench, Microsoft-backed start-up, Mistral, launched its newest (and biggest) model: Large 2, which builds on its predecessor and is supposedly on par with OpenAIās GPT-4o and Metaās Llama 3.1 (and better than Anthropics Claude 3.5 Sonnet), in performance, while delivering cost benefits, due to its smaller size.
š Key Points:
Large 2 has advanced multilingual capabilities (understanding languages like Arabic, Hindi, Chinese, Japanese, and Korean), and scored the same as Llama 3.1 in this area, in benchmark tests.
It also understands over 80 coding languages and has advanced math capabilities, outperforming Anthropics Claude 3.5 Sonnet, but coming a close 2nd to GPT-4o in both code generation and math benchmarks.
It follows instructions better, gives more concise responses, and, to reduce hallucinations, Mistral has fine-tuned it to be more discerning, so it will tell the user when it doesnāt know something, instead of fabricating an answer.
š¤ Why you should care: While Large 2 is open-source for non-commercial uses (unlike GPT-4o), users must pay if they want to use it commercially (unlike Llama 3.1, which is completely open-source), and it's also worth noting that, like Llama 3.1, it doesnāt have multi-modal capabilities, meaning it can just handle text, not images, audio, and video like GPT-4o can.
Which do you think is better in terms of performance vs cost? |
ā ā ā ā ā ā
Together with Augie Studio
Welcome to Augie Studio, an all-in-one video studio for businesses, creators, advertisers, and AI enthusiasts to create social videos, at scale.
Key features include:
Dynamic asset library: Use over 100M high-quality images from Getty Images, no licenses required.
Automatic content creation: Take advantage of AI-enhanced script writing and automated storyboarding to make unique video clips, efficiently.
Pro-level editing: Just click to apply transitions, captions, music, AI voiceovers, and effects, no tutorials needed.
PROMPT ENGINEERING
ā ā ā ā ā ā
Thursdayās Prompt: How to adapt a brand strategy with ChatGPT
Type this prompt into ChatGPT:
Help me understand how to adapt our brand strategy for different cultures and international markets
Results: After typing this prompt, you will get a brand strategy that allows you to maintain a consistent brand identity across different cultures and international markets.
P.S. Use the Prompt Engineer GPT by AI Tool report to 10x your prompts.
Finding the right candidate is challenging. Humanly.io revolutionizes hiring by automating screening, scheduling, and candidate engagement. Its AI screens resumes, coordinates interviews, answers questions 24/7, and transcribes interviews, highlighting key points for easy review.
For smaller-scale hiring, use tools like Fireflies or Otter.ai to record interviews, then integrate GPT for Sheets to analyze transcripts. Extract key details and get high-level overviews of each candidate, saving hours and improving decision-making.
BREAKING NEWS
ā ā ā ā ā ā
AI SEARCH
š Microsoft vs Google: AI search
Microsoft has previewed a new AI search featureāBing generative searchāwhich delivers AI-generated summaries in response to search queries, exactly like Googleās AI Overview feature.
Just a few users have the new feature, but Microsoft insists it āfulfills the intent of the userās query more effectivelyā than its AI-generated chat answers, which it launched in February.
Reports show that AI Overviews has reduced website visits by 25%, but Microsoft says test data proves it āmaintains the number of clicks to websitesā but will ālook closely at how AI search impacts visits to publishers.ā
RECOMMENDED RESOURCES
We read your emails, comments, and poll replies daily.
Hit reply and tell us what you want more of!
Until next time, Martin & Liam.
P.S. Donāt forget, you can unsubscribe if you donāt want us to land in your inbox anymore.
What did you think of this edition? |