AI Transformation Brief – September 2024

Author(s): Bill Wong

VOL 9 | SEPTEMBER 2024

INFO-TECH AI INSIGHTS

AI Transformation Brief

Featuring AI best practices and insights to enable our members to strategize, plan, develop, deploy, manage, and govern AI-based technologies and solutions.

Dividing line for the headline section.

In This Issue –› AI in the News | AI Research Highlights | Vendor Spotlight | Upcoming Events & Resources

AI IN THE NEWS

OpenAI introduces its next generation of models capable of advanced reasoning

Read Announcement

On September 12, 2024, OpenAI unveiled its latest models, o1-preview (for complex reasoning tasks) and o1-mini (cost-effective model for STEM fields – coding and mathematics), marking a significant leap forward in AI capabilities. Designed to excel in complex reasoning tasks, these models are poised to revolutionize how we approach challenging problems, especially in science, coding, and mathematics.

ANALYST ANALYSIS

The o1 family of models have been trained with reinforcement learning to perform complex reasoning. o1 appears to think before it responds. It takes time to produce its reasoning in a step-by-step manner and assess its response before responding to the prompt. In this way, o1 can improve its accuracy and mitigate bias and hallucinations. o1 outperforms GPT-4o in almost every reasoning category.

Note: 01’s pricing is significantly higher, at $15/1 million input tokens and $60/1 million output tokens, compared to GPT-4o at $5/1 million input tokens and $15/1 million output tokens.

California legislature passes comprehensive AI safety bill

Read News Article

On August 28, 2024, the California State Assembly passed the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act (SB-1047) with a 41-9 vote, marking one of the first significant legal frameworks in the US to regulate AI.

SB-1047 targets models that have been trained with more than 1026 floating-point operations (FLOPs) of compute and costs over $100 million to train. Note, the compute threshold specified is the same as what the White House used in its Executive Order for the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.

The Governor has until September 30 to sign the bill or veto it. If signed, the bill will become law by January 1, 2025.

ANALYST ANALYSIS

The goals of the legislation include balancing AI innovation and AI safety for the public, mitigating potential risks, establishing safety standards, increasing model transparency, and ensuring accountability from the organizations that build and deliver these frontier models.

New safety requirements would include a “kill switch” (to shut down the model), safety assessment requirement, third-party model testing, annual compliance certification, and penalties for noncompliance. In addition, it introduces protection for whistleblowers and gives the attorney general the power to seek civil penalties and monetary damages from developers and organizations that are building and deploying these models.

The legislation has sparked controversy in the AI community, and unfortunately much of the discussion is based on misinformation. This bill has been designed to regulate less than 10 corporations (OpenAI, Google, Meta, Anthropic, xAI, etc.) and their future models. The bill does not apply and will not regulate startups, the academic community, nor developers tinkering with models.

Microsoft to Rebrand Copilot

Read News Article, Read News Article (Subscription Required)

On September 16, 2024, Microsoft will deliver a Copilot event on LinkedIn hosted by Microsoft CEO Satya Nadella.

The company is expected to rebrand and ship new features to the service, including Copilot Pro, to give it a competitive advantage over ChatGPT Plus and Gemini Advanced.

ANALYST ANALYSIS

When a company rebrands a product, it is often to reset expectations and reposition its offering in the market. Often, the change is driven by failure to meet goals and expectations, usually with respect to adoption and sales. While thousands of organizations are assessing 365 Copilot around the world, very few have plans to deploy it across the enterprise. The most common reasons shared with me for not adopting 365 Copilot company-wide include Copilot’s inaccurate responses and slow response times as well as lack of a business case that justifies the costs. Mere product rebranding will not address these issues.

Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials

Read Press Release

On August 14, 2024, the FTC announced the long-awaited issuance of the Trade Regulation Rule on the Use of Consumer Reviews and Testimonials (Final Rule or Rule). This Rule will combat fake online reviews and testimonials by prohibiting their sale or purchase and allow the agency to seek civil penalties (fines up to $50,000 per violation) against knowing violators.

“Fake reviews not only waste people’s time and money, but also pollute the marketplace and divert business away from honest competitors,” said FTC Chair Lina M. Khan.

The Final Rule was published in the Federal Register on August 22 and will go into effect by October 21, 2024.

ANALYST ANALYSIS

The unanimous, bipartisan vote by the FTC is designed to protect consumers and will hold organizations and influencers accountable for any fake reviews and testimonials. The Final Rule prohibits:

Online fake or false consumer reviews, consumer testimonials, and celebrity testimonials. This includes fake reviews generated by AI or by reviewers that do not exist.
Buying positive or negative reviews
Using company insiders to write reviews
Positioning company-controlled review websites as independent sites
Using legal threats and/or other intimidation tactics to remove negative reviews
Misuse of fake social media indicators

While many are welcoming this legislation, it should be noted that third-party review platforms (e.g. Google, Yelp, Meta/Facebook, Tripadvisor) are not held directly accountable.

Wall Street Journal Interviews Info-Tech on Anthropic’s Latest Offering for Business Customers

Read News Article

On September 4, 2024, the Wall Street Journal reported that Anthropic’s latest offering, Claude Enterprise, has been released, targeting enterprise customers (as opposed to the consumer market).

New capabilities include larger context windows, integration with GitHub Copilot, and improved customization capabilities to improve alignment, safety, and security.

ANALYST ANALYSIS

Funded by AWS and Google, Anthropic’s Claude Enterprise offering is aimed squarely at OpenAI, Google Gemini, and other enterprise-class LLMs being marketed.

The performance and accuracy of Claude can rival/exceed GPT-4o (depending on the use case). What Claude required was an offering that was tailored for the business market to include improved security and administrative support. This enterprise-class offering continues to differentiate its performance with its implementation of responsible AI capabilities.

Dividing line for the headline section.

AI RESEARCH HIGHLIGHTS

(Re)Calibrate Your IT Workforce for AI — New!

Slide titled '(Re)Calibrate Your IT Workforce for AI' with an equation 'AI use cases + Strategy = Organizational Impact'. Below the equation is a diagram illustrating the variables.