AI Reshaping the Internet Landscape: From ChatGPT to Gemini

AI Chatbots Becoming More Accessible to the Public

How Large Language Models (LLMs) Function

Major tech companies like Microsoft (Copilot), Google (Gemini), and OpenAI (ChatGPT-4) are making AI chatbot technology more widely available to the general public. These AI tools operate as extensive autocomplete systems, trained to predict the next word in a given sentence based on statistical language properties. Rather than relying on a hard-coded database of facts, they generate plausible-sounding statements, which may not always be factual.

The Evolving AI Landscape

The AI landscape is rapidly evolving, with numerous developments and name changes occurring frequently. As these advancements unfold, The Zero Byte will keep you informed about the latest updates in the field.

Gemini’s AI Responses Now More Precise

Google’s Gemini chatbot has recently been updated to allow users to directly edit and fine-tune its responses within the chatbox. Starting March 4th, users can select a portion of text, provide instructions to Gemini, and receive an output that more closely aligns with their expectations. This feature is currently available in English on the Gemini web app.

ChatGPT’s Read Aloud Feature

OpenAI has introduced a new Read Aloud feature for ChatGPT, which can read its responses out loud to users in one of five voice options. This feature is available on both the web version and mobile apps (iOS and Android) of ChatGPT. Read Aloud supports 37 languages and automatically detects the language of the text it’s reading. It works with both GPT-4 and GPT-3.5 models, showcasing OpenAI’s multimodal capabilities.

Anthropic’s Claude 3 AI Models

Anthropic, an AI company founded by former OpenAI employees, claims that its latest Claude 3 family of AI models performs as well as or better than leading models from Google and OpenAI. Unlike earlier versions, Claude 3 is multimodal, capable of understanding both text and photo inputs.

Anthropic says Claude 3 will answer more questions

Anthropic Unveils Three New Claude AI Models

Introducing Claude 3 Haiku, Sonnet, and Opus

Anthropic has announced the release of three new AI models in the Claude 3 series: Haiku, Sonnet, and Opus. These advanced models boast improved capabilities in understanding longer instructions and providing more accurate responses. With increased context comprehension, Claude 3 can process and analyze a greater volume of information.

Opus Leads the Pack as the “Most Intelligent Model”

Among the three new models, Claude 3 Opus stands out as the largest and most sophisticated. Anthropic proudly declares Opus to be their “most intelligent model” to date. Both Opus and Sonnet are now accessible through claude.ai and the company’s API, while Haiku is slated for release in the near future.

Versatile Applications Across Various Domains

The Claude 3 series demonstrates remarkable versatility, with all three models capable of being deployed in a wide range of applications. These include chatbots, auto-completion systems, and data extraction tasks, showcasing the models’ adaptability to diverse use cases and industries.

Brave Introduces AI Browser Assistant on Android

Leo Makes the Leap from Desktop to Mobile

Brave, the privacy-centric browser, has expanded the reach of its AI assistant, Leo, by launching it on Android. Previously available only on desktop, Leo now joins the ranks of other mobile AI-connected browsers such as Edge and Arc (currently limited to iOS).

Empowering Users with AI-Driven Features

Leo offers a suite of powerful features, including summaries, transcriptions, translations, and coding assistance. However, Brave acknowledges the potential for Large Language Models (LLMs) to occasionally generate erroneous information, known as “hallucinations.”

Prioritizing User Privacy and Anonymity

In line with Brave’s commitment to privacy, the company assures users that their inputs are submitted anonymously through a reverse-proxy system and are not retained or used for training purposes. This approach aims to safeguard user data while still providing the benefits of AI-driven assistance.

Google Cloud Partners with Stack Overflow for Enhanced Coding Suggestions on Gemini

Seamless Integration for Developers

Google Cloud has formed a strategic partnership with Stack Overflow to provide developers with enhanced coding suggestions on the Gemini platform (not to be confused with the Gemini chatbot). This collaboration will enable developers using Gemini for Google Cloud to access Stack Overflow directly, with the new features set to roll out in the first half of 2024.

Leveraging AI for Improved Content Moderation and Engagement

Stack Overflow, which underwent significant layoffs last year amidst the surge in AI coding, will leverage Google’s AI services to streamline its content approval process and optimize forum engagement experiences. This partnership aims to harness the power of AI to enhance the platform’s efficiency and user experience.

The Evolution of Chatbots: From Eliza to ChatGPT

A 60-Year Quest for Human-Computer Interaction

The pursuit of seamless communication between humans and computers has been a driving force in the tech industry for nearly six decades. Many believed that cracking the code to natural language interaction would make computers more accessible, user-friendly, and enjoyable to use.

Eliza: The Pioneering Chatbot of the 1960s

The journey began in the 1960s with Eliza, a chatbot developed by MIT professor Joseph Weizenbaum. Designed to emulate a therapist, Eliza marked a significant milestone in enabling natural language conversation between humans and computers. Weizenbaum’s groundbreaking work laid the foundation for the chatbot revolution that would follow.

ChatGPT: The Latest Chapter in the Chatbot Story

Fast forward to the present day, and ChatGPT represents the latest iteration of this ongoing quest. The current AI chatbot revolution builds upon the decades-long efforts to bridge the gap between human and machine communication, offering unprecedented levels of natural language understanding and generation.

Google CEO Addresses Gemini AI Diversity Errors

Sundar Pichai: Errors are “Completely Unacceptable”

In an internal memo obtained by The Zero Byte, Google CEO Sundar Pichai addressed the recent controversy surrounding the Gemini AI’s generation of historically inaccurate images and text. Pichai acknowledged that the errors have “offended our users and shown bias,” deeming them “completely unacceptable.”

Pausing Image Generation and Working Towards a Solution

Last week, Google temporarily suspended Gemini’s image generation capabilities after it was discovered that the model produced racially diverse Nazi-era German soldiers, non-white US Founding Fathers, and inaccurately portrayed the races of Google’s own co-founders. While the company has apologized for “missing the mark” and is working to re-enable image generation in the coming weeks, Pichai’s memo marks the first time the CEO has widely addressed the issue.

Illustration by Cath Virginia / The Zero Byte

Wendy’s Betrays Spicy Nugget Lovers Everywhere

Wendy’s to Test Surge Pricing for Popular Menu Items

Image: Wendy’s

Picture this: you’re in the drive-thru queue at Wendy’s during peak lunch hours, eagerly anticipating your favorite spicy chicken nuggets. But as you pull up to place your order, the digital menu board reveals an unwelcome surprise – those coveted nuggets now cost an extra buck due to high demand. This could soon become reality as Wendy’s gears up to implement surge pricing for its most popular offerings during busy periods.

Wendy’s CEO Announces $20M Investment in Digital Menu Boards

In a recent earnings call, Wendy’s CEO Kirk Tanner unveiled plans to invest $20 million in rolling out digital menu boards across US locations by 2025. Along with this upgrade comes the introduction of “dynamic prices” that will fluctuate based on demand, akin to Uber’s surge pricing model which charges higher rates in busy areas.

Microsoft Expands Copilot GPTs for Fitness, Cooking, and More

Microsoft Copilot now offers a range of specialized GPTs to assist users with workouts, recipe discovery, vacation planning, and more. Corporate VP Jordi Ribas also teased the upcoming ability for users to create their own custom Copilot GPTs, currently in testing.

Microsoft Forges $2.1B Partnership with French AI Startup Mistral

Illustration: The Zero Byte

Microsoft has entered into a multiyear partnership with Mistral, a 10-month-old French AI company valued at €2 billion ($2.1 billion). The deal involves Microsoft taking a minor stake in Mistral and making its open and commercial language models available on Azure AI, becoming the second company after OpenAI to do so. The partnership will focus on developing and deploying next-gen large language models.

Google DeepMind CEO Addresses Gemini’s Controversial AI Images

During a keynote at the Mobile World Congress, Google DeepMind CEO Demis Hassabis acknowledged that Gemini’s model applied a range of people for images “too bluntly.” He announced that the paused photo generation feature is being fixed to provide a more historically accurate range of people and will return in a few weeks.

Google Clarifies Gemini’s “Embarrassing” AI-Generated Nazi Images

Illustration: The Zero Byte

Google has issued an explanation for the “embarrassing and wrong” images generated by its

Google Apologizes for Gemini AI’s Historical Inaccuracies

Tuning Issues Lead to Problematic Image Generation

Google has acknowledged that its Gemini AI tool produced “inaccurate historical” images due to tuning problems. In a blog post on Friday, Prabhakar Raghavan, Google’s senior vice president, explained that their efforts to ensure Gemini showed a diverse range of people failed to account for cases where diversity should not be applied.

“Our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range. And second, over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive.”

Gemini Generates Racially Diverse Nazis and Founding Fathers

Earlier this week, The Zero Byte and other sources discovered that Gemini was generating images depicting racially diverse Nazis and US Founding Fathers. This issue has raised concerns about the AI’s ability to accurately represent historical figures and events.

Microsoft’s PyRIT Tool Identifies Malicious AI Content

Microsoft has introduced PyRIT (Python Risk Identification Toolkit), an automated AI red teaming tool that can identify malicious content generated by AI systems within hours. PyRIT helps human evaluators focus on “hot spot” categories that might produce harmful prompt results.

Microsoft utilized PyRIT while redteaming its Copilot services, generating thousands of malicious prompts and scoring the responses based on potential harm in various categories. This process allows security teams to prioritize and address the most critical issues.

Google Partners with Reddit for AI Training Data

Google has entered into a new partnership with Reddit to access the platform’s data API for AI training purposes. This collaboration will provide Google with an efficient and structured way to utilize Reddit’s vast content corpus, enabling the company to train its AI models more effectively and display Reddit content in innovative ways across its products.

Google Apologizes for Missing the Mark with Gemini

Google has issued an apology for the “inaccuracies in some historical image generation depictions” produced by its Gemini AI tool. The company acknowledged that its attempts to create a “wide range” of results fell short, particularly in cases where diversity should not have been applied, such as with the US Founding Fathers or Nazi-era German soldiers.

“We’re aware that Gemini is offering inaccuracies in some historical image generation depictions. We’re working to improve these kinds of depictions immediately. Gemini’s AI image generation does generate a wide range of people. And that’s generally a good thing because people around the world use it. But it’s missing the mark here.”

Google is actively working to address these issues and improve Gemini’s performance in generating accurate historical depictions.

Microsoft’s Copilot Pro: A Month-Long Journey into the AI-Powered Future

Exploring the $20 Monthly Subscription

Last month, Microsoft unveiled Copilot Pro, a $20 monthly subscription that grants users access to AI-powered features within select Office apps. Additionally, subscribers receive priority access to the latest OpenAI models and enhanced image generation capabilities.

Over the past month, I’ve been putting Copilot Pro through its paces to determine if the $20 subscription is worth it for my daily needs and to assess the quality of the AI image and text generation across Office apps like Word, Excel, and PowerPoint. While some of the Copilot Pro features are currently underwhelming, others have proven to be genuinely useful improvements that I’m hesitant to part with.

OpenAI Employee Showcases Sora’s Video Collage Capabilities

An OpenAI employee recently demonstrated another impressive ability of the company’s new text-to-video generator, Sora. The AI can create stunning video collages, although the content in the upper right frame of the showcased example is somewhat perplexing.

OpenAI’s Struggle to Trademark ‘GPT’

The US Patent and Trademark Office (PTO) has rejected OpenAI’s application to register the word “GPT,” which stands for “generative pre-trained transformer.” The PTO stated that GPT is too general a term to register and could hinder competitors from accurately describing their products as GPTs.

In its application, OpenAI contended that GPT is not a descriptive word and that consumers would not “immediately understand” its meaning. However, the PTO’s decision suggests otherwise.

Canadian Court Holds Air Canada Accountable for Chatbot Misinformation

In a recent case, a Canadian court ruled that companies are responsible when their customer service chatbots provide false information to customers. The case involved a man who, while booking an Air Canada flight, inquired about a reduced bereavement rate. The chatbot assured him that a rebate was possible, but when he submitted the request, the airline refused to refund him.

In February of 2023, Moffatt sent the airline a screenshot of his conversation with the chatbot and received a response in which Air Canada “admitted the chatbot had provided ‘misleading words.'”

The man took Air Canada to court and emerged victorious, setting a precedent for corporate accountability in chatbot interactions.

Scientists Alarmed by AI-Generated Images in Peer-Reviewed Paper

Scientists are expressing grave concerns over a rat’s “dck” and several other nonsensical AI-generated images that were openly credited to Midjourney in a peer-reviewed science paper published by the Frontiers Journal this week. The gibberish annotations and grossly inaccurate images included in the paper highlight the risks that generative AI poses to the accuracy of academic research.

Frontiers has since responded by removing the offending paper, stating:

Our investigation revealed that one of the reviewers raised valid concerns about the figures and requested author revisions. The authors failed to respond to these requests. We are investigating how our processes failed to act on the lack of author compliance with the reviewers’ requirements.