Silicon Valley School’s newest tutor is ChatGPT

Hello folks, today we'll be discussing a Silicon Valley school officially welcoming an AI tutor and concerns regarding the safety of AI chatbots.

 As always if you enjoy reading our posts be sure to spread the word ! 

Here’s what we have lined up for you today - 

1. Silicon Valley School’s newest tutor is ChatGPT

2. AI models and security concerns

3. Adobe vs Midjourney

Silicon Valley School’s newest tutor is ChatGPT

OpenAI and Sam Altman believe that AI will create a 1:1 tutor for every learner in this world.  But many schools are limiting or even outright restricting the use of ChatGPT in the classroom. 

There is one school however in Silicon Valley that is embracing it. The Khan Lab School has introduced Khanmigo  built by their sister organisation Khan Academy. Teachers have expressed that students have been engaging actively with the tool and asking it doubts and questions. What’s novel is that the tool is personalised to each students’ learning needs based on the type of questions they ask. 

The tool is monitored by the teachers to ensure that students are not misusing it for purposes beyond their learning needs. While many are concerned about the need for safety guardrails, Khan Academy feels that it is taking a thoughtful approach as they improve the product. 

Calculators were once frowned upon in schools, today they are ubiquitous. The same could very well be the case for AI.

AI models and Security concerns

As impressive as AI chatbots are, they are exposed to vulnerabilities that bad actors have been taking advantage of. Because using these chatbots does not require any technical expertise, hackers are using something as simple as prompts to misuse these AI systems. 

Jailbreaking through prompting , scamming and phishing by generating fake content, and data poisoning are 3 ways that hackers are using to break into these systems. 

As these AI models and chatbots are deployed into products that are doing real life tasks like scheduling meetings, making reservations and replying to emails, putting up guardrails against such attacks is going to become more important. 

The problem however is that there is no one size fits all solution. With so many use cases already being explored there are way too many edge cases from which these systems can be attacked. 

This has partially been the reason why OpenAI has not released the details of their GPT-4 model and had an extensive safety report included with its launch. 

Safety at the cost of progress and vice versa is an important balance that all the key stakeholders will need to think about starting today. Here’s a little history lesson about a similar situation the world faced in the 1970s with DNA technology

Around the industry - 

1. Bill Gates says calls to pause AI won't 'solve challenges'

2. AI voice filter mimicking billionaire rapper Jay-Z raises concern as spectre of AI learning looms

3. Anthropic’s Claude is now available in Slack

4. Early comparisons of Adobe’s firefly model with Midjourney V5

5. Berkley just released Koala-13B another ChatGPT like chatbot built using smaller models

6. Towards Healthy AI: Large Language Models Need Therapists Too

Product Corner - 

There are way too many cool products to showcase just one !

• Kapa AI : ChatGPT for developer facing products 

• ColorMagic : Generate custom color palettes using AI  

• Sapling : Language model copilot for customer-facing teams respond twice as fast.

• Butternut AI : Create a website for your business in 20 mins

• Create : Describe your idea and see it become a webapp

• Fast Chat : ChatGPT style chat built using facing LLaMA

Adobe vs Midjourney

This is a pretty detailed thread with important caveats about competing approaches to generative art !