AI Pulse Edition #6: Latest AI News Updates for the Developer Community

Welcome back to AI Pulse—your human curated roundup of the last two weeks of AI news, featuring a peek into the exciting work our community is creating and a chance to share your opinion on recent AI news.

BREAKING: Greg Brockman returns to OpenAI!

We’re launching a new Community Spotlight series this edition that highlights projects of community members! First up is Convo by @stevenic, a simple project that’s been generating some great conversations!

Do you have a project you’d like to have highlighted in a future Community Spotlight? Leave a comment in the thread to let us know!

Also new this edition, we’ve got coverage of OpenAI’s DevDay with a behind-the-scenes view from community reporter @Platypus, who shares their experience at the London event.

Last but not least, we have a list of the biggest AI news of the last two weeks in a variety of spheres.

Here’s to another great edition of AI Pulse. Happy reading, and let’s keep building the future together!

Brought to you by the AI Pulse team: @jr.2509 @vb @platypus @trenton.dambrowitz @dignity_for_all @PaulBellow

P.S. Interested in joining our team? Get in touch—we’d love to have you on board!


Table of Contents

1. Community Spotlight
2. Technology Updates
3. Infrastructure
4. Government & Policy
5. Legal Matters
6. Business & Economy
7. Research
8. Arts & Entertainment
9. Dev Alerts
10. Special Events++


1. Community Spotlight

In this edition of the newsletter we continue our community spotlight. We are varying things up a bit this time, by putting a spotlight on a tool developed by one of our community members, and by getting another community member to give us a summary of the most recent OpenAI DevDay in London.

Community Tool - Convo

Convo from our community member @stevenic, is a high-level, natural language-based programming language optimized for use with Large Language Models (LLMs). It simplifies coding by allowing users to write in plain, intuitive language rather than traditional syntax, making programming more accessible. Convo functions as pseudocode, enabling developers to outline program logic before converting it into specific languages like Python.

It offers a number of key benefits, including:

  1. Predictability and Control: Starting with Convo allows users to define program features more clearly.

  2. Ease of Use: Convo’s syntax resembles natural conversation, using flexible, descriptive commands.

  3. Multilingual Support: Convo programs can be written in various languages, improving accessibility.

  4. Efficient AI Code Generation: Convo leverages an LLM’s strengths in pseudocode generation to reduce errors in final code output.

Check out Convo today, or reach out to @stevenic directly!

London DevDay Insights

It was a rather warm late-October morning. After managing to ignore the construction noise on Beech St, I finally entered The Brewery off Chiswell St - an elegant 18th century brewery that now played host to the first ever European OpenAI DevDay event. After showing my confirmation email at three different checkpoints, I finally entered the venue and was warmly greeted by the organizing team. The event was professional and well organized, the swag was spot on (I am wearing that hoodie even as I write this!), and the crowd was a wonderful mixture of serious AI builders and everlasting dreamers. I don’t know if it was just purely coincidental, but I spoke to four different LegalTech teams that day - using AI in legal just a few years ago was unthinkable!

I could go on the whole day, but instead I will try to summarize my key takeaways that are based on the keynotes, all the demos and the AMA with Sama:

  • No new models were announced during the event. However the most significant practical announcement was regarding Realtime API - OpenAI implemented caching optimizations so that the price of Realtime API is lowered by 2-5x on average.

  • A number of practical deep-dives were held throughout the day, including a session on structured outputs, which leverage token masking and constrained sampling; as well as a session on distillation, where the aim is to either focus on low-precision but broadly general tasks, or very narrow tasks.

  • o1 model family will “soon” be promoted from preview, and will feature multi-modality as well as function calling and structured outputs - the exact date however was not announced.

  • o1 and the new paradigm of “reasoning models” is of great importance and a strategic direction for OpenAI. o1 family of models are “on a quite steep trajectory of improvement” and OpenAI’s recommendation to the community is to prepare for these rapid improvements. Instead of building products and companies that try to patch or overcome the existing gaps and limitations, OpenAI’s recommendation is to build products that will automatically benefit from future model improvements.

  • Using o1 for coding and low-code application development was very pronounced throughout the event. A number of demos were shown including building a food ordering agent app for iOS that leverages Realtime API; as well as an iPhone app that controls a DJI drone in realtime. Sam also emphasized that software development is where large gains have clearly been seen, and one his favorite companies/apps is Cursor.

I’d also like to take this opportunity to thank OpenAI for having me, and all the people involved in putting together a wonderful event. And wish you the best of luck at the upcoming DevDay in Singapore! @platypus

2. Technology

Recent AI advancements are transforming industries, from design and healthcare to web search and complex task automation. Google and OpenAI continue to push AI boundaries, with new tools for natural conversation, automated coding, and real-time search integration. TIME’s Best Inventions list highlights the growing impact of AI on daily life, including intuitive design tools, optimized surgeries, and secure digital services. Together, these developments showcase AI’s expanding role in both creative and operational spheres.

Pushing the frontiers of audio generation

Google DeepMind has advanced audio generation with a new speech model capable of creating realistic, multi-speaker dialogue. Their latest technologies, like NotebookLM Audio Overviews and Illuminate, turn complex content into engaging, AI-driven conversations, enhancing accessibility. Powered by innovations like SoundStream and AudioLM, the model can generate high-quality, 2-minute dialogues in under 3 seconds. This technology, safeguarded by watermarking, promises new possibilities in digital interactions and learning experiences.

Source: Google DeepMind

Google Now Uses AI for Over 25% of Its Code, CEO Says

Google and Alphabet CEO Sundar Pichai highlighted strong Q3 performance, with growth across Search, Cloud, and YouTube driven by the company’s commitment to AI innovation. AI-powered tools are boosting internal productivity, with over a quarter of Google’s new code generated by AI, while advancements in Search and Cloud show increased user engagement and deeper business adoption. YouTube and Waymo continue to grow, with record-breaking revenue and expansion in autonomous driving.

Source: Alphabet

Time Magazine’s ‘Best Inventions of 2024’ Highlights Top AI Innovations

TIME’s 2024 Best Inventions list highlights several transformative AI-powered innovations making waves across industries. Canva Magic Studio stands out as an intuitive design suite, enabling users to create images, write content, and design presentations effortlessly. Content Credentials offers a solution for combating misinformation by embedding metadata into images, ensuring their authenticity and traceability. In healthcare, EXeX ExperienceX optimizes operating theaters through AI, enhancing surgical efficiency and patient outcomes. The financial sector benefits from Column Tax, an app that simplifies tax filing by providing personalized AI-driven tax advice. Finally, Diia revolutionizes government services by enabling secure online voting and digital identification through AI, promoting accessibility and security in public services.

Source: TIME

OpenAI Introduces ChatGPT Search for Enhanced Web Search Capabilities

OpenAI has introduced a new ChatGPT search feature that enhances web search capabilities directly within ChatGPT. Available to Plus and Team users, it combines the ease of a natural language interface with access to up-to-date information like sports scores, news, and financial data. Users can initiate searches manually or ChatGPT can decide based on the query, with source links for deeper exploration. This feature, powered by partnerships with data providers, integrates across ChatGPT’s desktop, mobile apps, and the web, with plans to roll out to all users in the coming months, including enterprise and educational sectors.

Source: OpenAI

Black Forest Labs Unveils Enhanced Image Generation with FLUX1.1 Pro

Black Forest Labs has unveiled new features for FLUX1.1 [pro], enhancing its image-generation capabilities with “Ultra” and “Raw” modes. Ultra Mode allows users to generate images at up to 4MP resolution in just 10 seconds, maintaining prompt accuracy and operating 2.5 times faster than similar models at a competitive rate of $0.06 per image. Raw Mode, aimed at creators who want a more authentic look, offers a natural aesthetic that mimics candid photography, enhancing realism, especially in human subjects and nature scenes. Both modes are available now through the FLUX1.1 [pro] API, catering to high-resolution and authentic imaging needs.

Source: Black Forest Labs

Microsoft's Magentic-One: A Generalist Multi-Agent System for Complex Tasks

Microsoft Research has introduced Magentic-One, a generalist multi-agent system designed to handle complex, open-ended tasks across diverse domains, offering significant advancements in agentic AI. This system operates through an Orchestrator agent that coordinates four specialized agents—WebSurfer, FileSurfer, Coder, and ComputerTerminal—to perform tasks autonomously, such as browsing the web, managing files, and executing code. Magentic-One, built on the Microsoft AutoGen framework, achieves competitive performance on agentic benchmarks and is available open-source to encourage collaborative development. Although powerful, the system’s deployment emphasizes safety, incorporating tools like AutoGenBench for rigorous testing to mitigate risks associated with autonomous AI systems.

Source: Microsoft Research

3. Infrastructure

In infrastructure, AI and clean energy initiatives are taking center stage, especially as data centers increase energy demand. The White House recently convened experts to explore how AI and software could speed up clean energy grid integration, aiming to lower costs and reach climate goals. Efforts include a new AI-driven program from the Department of Energy to streamline application processes for clean energy projects. Meanwhile, companies like Amazon, Meta, and Microsoft are exploring nuclear power to fuel data centers, though regulatory hurdles have delayed Amazon and Meta’s plans due to environmental and reliability concerns, highlighting the challenges of meeting high energy demands sustainably.

Readout of White House Discussion on AI and Advanced Software Solutions to Accelerate Clean Energy Grid Integration

The White House Task Force on AI Datacenter Infrastructure recently gathered experts to discuss how AI and software solutions can expedite clean energy grid integration. The Biden-Harris Administration emphasized AI leadership and clean energy initiatives as national priorities, with goals to maintain low electricity costs and achieve climate targets. Key discussions included tackling the backlog of interconnection projects and employing AI to streamline grid integration, as highlighted by Secretary of Energy Jennifer Granholm and other senior officials. Additionally, the Department of Energy announced a forthcoming AI-driven program to expedite application processing for clean energy projects, building on earlier efforts under the Interconnection Innovative e-Xchange (i2X) program.

Source: The White House

Regulators Deliver Successive Blows to Amazon and Meta’s Nuclear Power Ambitions

Amazon, Meta, and Microsoft are increasingly turning to nuclear power to meet the rising energy demands of their data centers driven by AI and cloud computing. However, Amazon and Meta have recently encountered regulatory setbacks. Meta’s AI data center project near an existing nuclear plant faced delays due to environmental concerns, including the discovery of a rare bee species. Meanwhile, Amazon’s planned data center adjacent to Pennsylvania’s Susquehanna nuclear plant was hindered by a Federal Energy Regulatory Commission (FERC) decision, which cited potential reliability issues for other grid users. Microsoft’s nuclear project at Three Mile Island is still progressing, illustrating the ongoing regulatory complexities around nuclear power for hyperscale data centers.

Source: TechCrunch

4. Government & Policy

In government and policy, China’s adaptation of Meta’s Llama AI model for military use underscores the complexities of regulating open-source technology. Researchers linked to the Chinese military have reportedly customized the Llama model to create “ChatBIT,” a tool for intelligence and operational support, despite Meta’s restrictions against military applications. Meanwhile, the Republican Party’s recent platform indicates a shift in U.S. AI policy, as it plans to repeal existing executive orders that are perceived as restrictive and impose new regulations promoting AI innovation aligned with free speech and human flourishing.

The stance of the Republican party and president-elect Trump toward AI

The U.S. presidential election is finally over, and from the Republican Party’s platform, we can glimpse what lies ahead for AI regulation in the USA over the coming years. Though brief and somewhat lacking in detail, the section on AI is, in part, very clear:

“We will repeal Joe Biden’s dangerous Executive Order that hinders AI innovation and imposes radical left-wing ideas on the development of this technology. In its place, Republicans support AI development rooted in free speech and human flourishing.”

Source: The American Presidency Project

Chinese researchers develop AI model for military use on back of Meta’s Llama

Chinese researchers linked to the People’s Liberation Army (PLA) have reportedly adapted Meta’s open-source Llama AI model to create a military tool named “ChatBIT.” This AI model, based on Llama’s 13B language model, is intended for intelligence gathering and operational support. Despite Meta’s restrictions against military usage, the PLA’s use of the model highlights challenges in enforcing such policies for open-source AI. The Pentagon is monitoring these developments closely amid U.S. concerns over China’s growing AI capabilities. Meanwhile, Meta defends open innovation, emphasizing that China’s broader investments in AI remain a larger factor in the global competition.

Source: Reuters

5. Legal Matters

In a recent legal case, Raw Story Media, Inc. v. OpenAI Inc., the District Court for the Southern District of New York dismissed the claims against OpenAI. Judge Colleen McMahon granted the motion to dismiss and denied the plaintiffs leave to amend their complaint, though they may refile if they can show that amendments would be valid. This decision resolves the motion at Docket No. 68 in favor of OpenAI, streamlining the ongoing litigation.

Raw Story Media, Inc. v. OpenAI Inc. Dismissal in District Court

In the case Raw Story Media, Inc. v. OpenAI Inc., the District Court for the Southern District of New York granted OpenAI’s motion to dismiss on November 7, 2024. Judge Colleen McMahon’s decision denied the plaintiffs’ request for leave to amend their complaint, stating it could be renewed with a proper motion and proposed amendment showing that it would not be futile. This dismissal removes the motion at Docket No. 68 from the list of open motions.

Source: CourtListener

6. Business & Economy

Major corporations are making strategic moves to integrate and brand their AI innovations. Disney has set up a unit to manage AI and augmented reality across its entertainment and theme parks, aiming to transform customer experiences. OpenAI has acquired the domain Chat.com, redirecting it to its chatbot without altering branding. Microsoft is planning a rebrand of its Windows AI features as “Windows Intelligence,” unifying various tools under a single identity. Wendy’s is using Palantir’s AI to enhance supply chain efficiency, especially during high-demand promotions, demonstrating AI’s role in optimizing fast-food logistics.

Walt Disney Forms Business Unit to Coordinate Use of AI, Augmented Reality

Walt Disney has established the Office of Technology Enablement, a new unit to oversee the use of AI and mixed reality across its film, television, and theme park divisions. Led by Jamie Voris, Disney’s CTO, this team will ensure emerging tech projects align with the company’s broader strategy. Disney aims to leverage AI and XR (extended reality) to transform consumer experiences, while also navigating associated risks. The team, expected to grow to around 100 members, is part of Disney’s larger effort to integrate augmented and virtual reality into its offerings, capitalizing on trends in immersive tech to enhance its theme parks and in-home experiences.

Source: Reuters

OpenAI Acquires Chat.com, Redirects to ChatGPT

OpenAI recently acquired the domain Chat.com, which now redirects to ChatGPT. The historic domain, initially registered in 1996, was reportedly purchased last year by HubSpot co-founder Dharmesh Shah for $15.5 million, making it one of the most expensive domain acquisitions on record. Shah confirmed he sold the domain to OpenAI, likely in exchange for OpenAI shares, though the exact sale amount remains undisclosed. Despite the acquisition, this move doesn’t suggest a rebranding for ChatGPT, as OpenAI is not hosting the chatbot directly on Chat.com.

Source: TechCrunch

Microsoft Considers Rebranding Copilot to Windows Intelligence

Microsoft is reportedly considering rebranding its AI features in Windows 11 as “Windows Intelligence,” echoing Apple’s recent move with “Apple Intelligence.” This potential rebrand, based on references found in the latest Windows 11 builds, would bring together AI-powered tools like Copilot under a unified brand. This comes after significant changes to Copilot, which included the addition of new features but received user criticism for a degraded experience. Microsoft has not officially commented on the rebrand, but the move indicates a continued focus on integrating AI features cohesively within Windows.

Source: Windows Central

Wendy's $1 Frosty Deal Made Easier with AI from Palantir

Wendy’s is collaborating with Palantir Technologies to manage demand surges and supply chain logistics, especially around promotions like its popular $1 Frosty. Using Palantir’s AI capabilities, Wendy’s can predict potential shortages, ensuring the fast-food chain is prepared to meet customer demand without disruption. The partnership reflects Wendy’s strategy to use advanced technology to compete with larger rivals and may eventually allow for AI-driven ordering processes without human intervention. Wendy’s supply chain executive highlighted that this integration with Palantir’s technology could transform their operational approach.

Source: Bloomberg

7. Research

Research in AI is advancing toward more sophisticated and practical applications. Meta’s FAIR team has made breakthroughs in touch perception, dexterity, and human-robot interaction, aiming to bring Advanced Machine Intelligence (AMI) closer to reality. Their new tactile sensing models and human-robot interaction frameworks, supported by partnerships with GelSight Inc and Wonik Robotics, set a foundation for AI to interact more naturally in physical environments. Meanwhile, AI firms, including OpenAI, are moving beyond scaling with data and computation by using “test-time compute” to enhance AI responses through multi-step reasoning. This shift could alter AI infrastructure needs, steering demand toward inference optimization and potentially diversifying the hardware market.

Meta (FAIR) - Advancing Embodied AI Through Touch Perception, Dexterity, and Human-Robot Interaction

Meta’s Fundamental AI Research (FAIR) team has announced new advancements in robotics, focusing on touch perception, dexterity, and human-robot interaction to drive progress toward Advanced Machine Intelligence (AMI). Key releases include Meta Sparsh, a touch representation model that enables versatile tactile sensing, and Digit 360, an advanced tactile sensor with human-level precision, to enhance AI’s understanding of the physical world. Meta also introduced Meta Digit Plexus, a platform for integrating tactile sensors on robot hands. Strategic partnerships with GelSight Inc and Wonik Robotics will commercialize these innovations, making them accessible to researchers globally. Additionally, Meta introduced the PARTNR benchmark, a standardized framework for assessing AI’s ability to collaborate with humans, paving the way for AI models that can function as reliable partners in complex, real-world scenarios. These developments aim to advance robotics research and foster new AI applications across industries.

Source: Meta

New Path to Smarter AI due to Current Limitations

As AI companies like OpenAI confront limitations in scaling large language models, they are exploring alternative techniques to achieve smarter AI. Instead of focusing solely on scaling with more data and computational power, researchers are increasingly adopting “test-time compute” methods, which enhance model inference by allowing AI to evaluate multiple responses in real time before choosing the best one. OpenAI’s recent “o1” model exemplifies this approach by employing multi-step reasoning similar to human thought processes. This shift could reshape the demand for AI infrastructure, with a growing focus on inference clouds over traditional training clusters, potentially affecting the dominance of Nvidia’s chips in the market.

Source: Reuters

8. Arts & Entertainment

The Arts & Entertainment sector is seeing significant impacts from AI, with both practical and ethical implications. In Poland, Off Radio Kraków’s decision to replace human journalists with AI hosts has stirred public outrage, especially given the station’s government funding. This move has sparked a broader conversation on AI’s role in replacing jobs in creative fields. Meanwhile, in the art world, an AI-created portrait of Alan Turing by the humanoid Ai-Da Robot sold for over $1 million at Sotheby’s, marking a milestone as the first humanoid robot artwork to reach such a price. This sale highlights the growing intersection of AI and art, prompting debate on AI’s influence and ethical considerations within creative industries.

AI Replaces Human Journalists at Poland's Off Radio Kraków

Off Radio Kraków, a government-owned radio station in Poland, has sparked controversy by replacing all human journalists with AI hosts, igniting public outrage and a national debate. Former journalist Mateusz Demski, along with thousands of supporters, argues that this move threatens jobs in media and creative industries. The station’s three AI hosts, “Kuba,” “Emi,” and “Alex,” cover topics such as technology, pop culture, and social issues, but critics claim that the decision to replace staff with AI is especially problematic for a taxpayer-funded station. This development highlights growing concerns about AI’s impact on employment in creative fields, with Polish government officials also expressing reservations about the ethical boundaries of AI adoption.

Source: ZME Science

AI Portrait of Alan Turing by Ai-Da Robot Fetches Record Price at Sotheby’s

An AI-generated portrait of Alan Turing, titled “A.I. God,” created by the humanoid Ai-Da Robot, recently sold at Sotheby’s for $1,084,800, far surpassing its initial estimated value. This record-setting sale is notable as the first artwork by a humanoid robot to achieve such a price at auction, reflecting the rising intersection between AI and the art world. Ai-Da’s portrait series of Turing, known as the father of artificial intelligence, serves as a prompt for viewers to contemplate AI’s advancing role and its ethical implications. The director of Ai-Da Robot Studios emphasized that the sale marks a transformative moment in visual arts, posing questions about the agency and power of AI in creative spaces.

Source: BBC

9. Dev Alerts

OpenAI has continued to introduce new features to enhance AI applications across various domains as shown in the Docs and the OpenAI Cookbook.

Steering Text-to-Speech for more dynamic audio generation

OpenAI’s traditional TTS APIs don’t have the ability to steer the voice of the generated audio. For example, if you wanted to convert a paragraph of text to audio, you would not be able to give any specific instructions on audio generation.

With audio chat completions, you can give specific instructions before generating the audio. This allows you to tell the API to speak at different speeds, tones, and accents. With appropriate instructions, these voices can be more dynamic, natural, and context-appropriate.

Source: OpenAI Cookbook

Predicted Outputs: Minimize response latency

Predicted Outputs enable you to speed up API responses from Chat Completions when many of the output tokens are known ahead of time. This is most common when you are regenerating a text or code file with minor modifications. You can provide your prediction using the prediction request parameter in Chat Completions.

Predicted Outputs are available today using the latest gpt-4o and gpt-4o-mini models. Read on to learn how to use Predicted Outputs to reduce latency in your applicatons.

Source: OpenAI API Documentation

10. Special Events ++

OpenAI’s recent AMA on Reddit covered several developments, including a focus on refining the o1 model series rather than releasing GPT-5 this year, with plans for future convergence of the GPT and o1 series. They’re working on expanding context windows and enhancing multimodal capabilities, like text-to-image generation, in GPT-4o. OpenAI also highlighted reduced inference costs, ongoing efforts to lower hallucinations, improved multilingual performance, and exploration of NSFW content controls.

For more (and to discuss), check out the AMA summary thread!

13 Likes

Thank you to all the contributors @jr.2509 @PaulBellow @vb @trenton.dambrowitz @dignity_for_all !

What do you like about it? What should be improved? Would you like to help and contribute? Please let us know!

3 Likes

Thanks for sharing your Dev Day experience!

We’re definitely looking for more people interested in helping… or spotlighting community projects!

2 Likes

Thanks for putting this together, great work!

3 Likes

Thank you @andy ! If you have any improvement suggestions or would like to contribute, please let us know!

3 Likes

Looking for a good dev spotlight for next issue too!

1 Like

Thanks for the call out! I’ve been enjoying these posts :slight_smile:

3 Likes

Here’s one : Request for improvement

1 Like

Thanks!

Our AI Pulse team is getting smaller, but we’re pushing forward. (We’re starting to build out tools to help put it together every week…If anyone’s interested in joining the team, let us know! )

We’ve got a good Community Spotlight for AI Pulse #7, but if anyone has seen anything noteworthy, please pass it along.

1 Like

Thanks @SweetT, will bring it to the Pulse team!

1 Like

New issue coming soon!

2 Likes

Last minute edits… We’ve got a banger this time…

1 Like