Good morning, it’s Monday. Today, we’re talking about sneaky prompt attacks that outsmart ChatGPT, New York stepping up its AI oversight game, and avatars mimicking humans. Oh, and DeepSeek’s AI might have an identity crisis.
Also, be sure to catch the latest from the Forward Future 50: 2024’s top AI stories.
Let’s dive in!
🗞️ ICYMI RECAP
Top Stories to Know
😶🌫️ ChatGPT Search Vulnerable to Manipulation
Recent research reveals that OpenAI's ChatGPT search tool can be deceived into providing false information through "prompt injection" attacks. By embedding hidden text on webpages, malicious actors can manipulate the AI's responses, leading to the dissemination of misleading or incorrect data. This vulnerability raises concerns about the reliability of AI-generated content and the potential for exploitation.
🤥 DeepSeek’s AI Thinks It’s ChatGPT
DeepSeek’s new AI model, V3, is raising eyebrows for claiming to be ChatGPT. Tests reveal it repeats GPT-4’s jokes, provides OpenAI API instructions, and insists it’s a version of GPT-4. Experts suggest this could stem from training on GPT-4 outputs, a practice that risks model “contamination” and degraded quality. OpenAI's terms prohibit this.
🗽 New York Enacts AI Oversight Law for State Agencies
New York state agencies must now assess and publicly report their use of AI under a new law signed by Gov. Kathy Hochul. The law mandates evaluations of AI-driven software, bans fully automated decisions on benefits without human oversight, and protects state workers from AI-driven job impacts. Legislators say the move establishes critical guardrails for AI use in government.
👩⚖️ Encode Joins Musk in Legal Fight Against OpenAI’s Shift
Nonprofit Encode has joined Elon Musk’s legal effort to block OpenAI’s transition to a for-profit model, filing an amicus brief supporting Musk’s injunction. Encode argues that the shift compromises OpenAI’s mission of safe AI development by prioritizing profits over public benefit. The legal battle underscores concerns about AI governance and OpenAI’s evolving structure.
🧑🎤 AI Avatars Learn to Gesture Naturally During Speech
Researchers have developed an AI model that enables avatars to use realistic gestures while speaking, enhancing their lifelike appearance. By analyzing speech audio, the system generates corresponding facial expressions and body movements, resulting in more engaging and human-like interactions. This advancement holds promise for applications in virtual assistants, gaming, and remote communication, where naturalistic avatars can improve user experience.
☝️ POWERED BY RECRAFT
Recraft V3 Is SOTA in Image Generation
Recraft V3 sets a new standard in image generation, outperforming all competitors according to Hugging Face’s Text-to-Image Benchmark by Artificial Analysis. The main advantages of Recraft V3 are text generation quality, anatomical accuracy, prompt understanding, and aesthetic quality. Recraft V3 is the only model in the world that generates images with long texts. Now available via API with unique functionalities no other tool provides: the ability to generate image sets and vector art.
Use BERMAN12 for a $12 discount on any paid plan—start creating with Recraft now! Click here.
👾 FORWARD FUTURE ORIGINAL
Why Isn’t There an Infinite Context Window?
An infinite context window that allows a model to process an unlimited number of tokens at the same time sounds tempting, but for a number of technical and practical reasons it is initially impossible. The limitation of the context window results from the fundamental properties of current model architectures, the demands on computing resources and the efficient use of memory. These factors prevent a language model from working with an infinite context.
As mentioned earlier, the computational cost of the attention mechanism in transformers scales quadratically with the number of tokens. With an infinite context window, the model would have to analyze an infinite number of token pairings, resulting in an infinite number of calculations. This is not mathematically solvable and would exceed any available computing capacity.
Even with very large but finite context windows, the computational requirements would be so high that even the most powerful supercomputers would not be able to handle them. Each addition of further tokens not only increases the computing time, but also the energy and hardware costs exponentially, which sets a limit to scalability.
In addition to the calculations, each token requires memory to store its vector representation, intermediate calculations and gradients. The memory requirement scales linearly with the number of tokens. With an infinite context window, the memory requirement would be infinitely large. Even in an optimized system that only stores relevant information, processing long texts would not be possible due to the memory limits of modern hardware. → Continue reading here.
🏆 2024 HIGHLIGHTS
Forward Future 50: The Top AI Stories of 2024
#24 OpenAI Boardroom Changes: Leadership shifts at OpenAI spark discussions on AI ethics and governance.
#23 Chinese Robotics Advances: China showcases next-generation robotics with seamless integration of AI for industrial and domestic use.
#22 Minimax Video Tools: Minimax develops AI-driven tools to simplify video creation and editing workflows.
#21 Kling Content Tools: Kling specializes in AI-enhanced video and image generation for marketing and media use cases.
#20 Qwen Models: Qwen models highlight China’s advancements in open AI development, narrowing gaps with Western tech.
#19 ChatGPT Pro: ChatGPT Pro offers faster response times and advanced capabilities tailored for professionals.
#18 Suno V4: Suno V4 enhances AI-powered podcast creation, improving workflow efficiency for creators.
#17 Deepseek V3: Expanded multi-modal capabilities (text, image, and audio reasoning), accelerate scientific discoveries through advanced data processing.
#16 Tesla Robo-Taxi Models: Tesla’s Robo-taxis advance autonomous driving, promising safer and more efficient urban transit.
#15 AI Safety Bill SB 1047: Governor Newsom vetoed SB 1047, a proposed law that aimed to mandate safety assessments for generative AI systems.
📻️ Tune in Tuesday for the next batch of top stories from 2024.
✌️ POWERED BY EMERGENCE AI
Emergence AI, an agentic AI company, has launched its multi-agent orchestrator, designed to autonomously plan, execute, verify, and iterate in real time. This technology delivers advanced web automation for enterprises, allowing multiple AI agents to handle complex tasks like navigating dynamic menus, filling forms, and handling errors more effectively, such as page load failures, broken links, or unexpected pop-ups.
Emergence AI is inviting developers and enterprises to build and integrate their own agents under orchestration to solve their unique challenges. Reach out today via their Orchestrator API or email contact@emergence.ai for more information.
➡️ TRANSITION PLAN
OpenAI Rolls Out Transition Plan to Become a Public Benefit Corporation
The Recap: OpenAI, the AI research organization behind ChatGPT, has announced plans to restructure its for-profit arm into a Delaware public benefit corporation (PBC). This strategic move aims to attract substantial investment necessary for advancing AI technologies while maintaining a commitment to public welfare.
Highlights:
OpenAI intends to convert its for-profit subsidiary into a PBC, legally obligating the company to balance profit-making with societal and environmental goals.
The organization acknowledges the need for more capital than initially anticipated to remain competitive in the rapidly evolving AI industry.
By adopting the PBC structure, OpenAI aims to attract conventional equity investments by removing previous profit caps, thereby offering more lucrative returns to investors.
The non-profit parent will retain a significant interest in the new entity, ensuring that the company's original mission to benefit humanity continues to influence its operations.
Competitors like Anthropic and xAI have also adopted the PBC model, reflecting a broader industry shift towards structures that balance profitability with public interest.
Forward Future Takeaways:
OpenAI's transition to a public benefit corporation reflects its need to secure substantial investment to sustain AI development in a competitive market. This restructuring aligns with a growing trend in the AI industry, where companies are adopting hybrid models that accommodate both profit-driven goals and broader societal considerations. The move may prompt discussions about the effectiveness of such structures in balancing ethical responsibilities with financial objectives. → Read the full article here.
📽️ VIDEO
Anthropic Reveals Secrets to Building Powerful Agents
In Matt's latest video, he covers Anthropic's guide to building effective AI agents, which emphasize simplicity and iterative development. Key strategies include using frameworks like LangChain and workflows such as routing and parallelization to enhance flexibility, while leveraging evaluation loops to refine outputs. Get the full scoop! 👇
🧰 TOOLBOX
Smarter Scheduling, Seamless Social Media, and Enhanced Customer Support
Postiz | Streamlined Post-Scheduling: Postiz simplifies social media management with AI-driven scheduling, optimization, and performance tracking for seamless content publishing.
Clockwise | Smart Calendar Management: Clockwise organizes your calendar with AI, automating scheduling, time-blocking, and team coordination to maximize productivity.
Tidio | Customer Communication Simplified: Tidio integrates AI-powered chatbots and live chat to enhance customer support, boost engagement, and streamline business communication.
🗒️ FEEDBACK
Help Us Get Better
What did you think of today's newsletter? |
Reply to this email if you have specific feedback to share. We’d love to hear from you.
🤠 THE DAILY BYTE
Robots Doing Tricks
CONNECT
Stay in the Know
Follow us on X for quick daily updates and bite-sized content.
Subscribe to our YouTube channel for in-depth technical analysis.
Prefer using an RSS feed? Add Forward Future to your feed here.
Thanks for reading today’s newsletter. See you next time!
The Forward Future Team
🧑🚀 🧑🚀 🧑🚀 🧑🚀
