AI
Trending Stories
Google’s Experimental Gemini Model Tops the Leaderboard, But Stumbles in My Tests
Arjun Sha
View quick summary
Google's upcoming Gemini-exp-1114 model has topped the LMArena leaderboard, outranking OpenAI and Anthropic models. However, in my reasoning tests, the model fails to answer tricky questions. In my book, it's still behind OpenAI's o1 models.
In the AI field, it was long believed that scaling and making larger models would lead to greater intelligence. However, Ilya Sutskever now says that LLM scaling has reached its limits. OpenAI, Google, and Anthropic are now focused on inference scaling to improve model performance.
ChatGPT Search is a promising alternative to Google Search, but it doesn't cover all types of search queries. For informational queries, ChatGPT Search is good, but for local, navigational, and shopping queries, it falls short.
Perplexity is working on an AI-powered shopping tool to make online shopping hassle-free. It searches the web to find the best product based on your prompt and you can quickly place the order without leaving the platform. It seems Perplexity is trying to compete with Google Shopping.
You can move away from Google and switch to ChatGPT Search to find the latest information from the web. OpenAI has released an official ChatGPT Search Chrome extension that replaces Google with ChatGPT Search. You can install the extension on any Chromium-based browsers.
Similar to ChatGPT's Code Interpreter which runs Python code, users can now run JavaScript code in Claude, and it runs in the browser itself. It can come in handy for performing data visualizations, number crunching, and more. Users can also use it to run various mathematical calculations.
Beebom Staff
Those looking for an AI video generator can check out Wondershare Virbo which delivers studio-quality AI videos in minutes. You can generate a script using AI and add natural speech using the text-to-speech model. It offers a friendly interface and you can even use it on your desktop and smartphone.
Microsoft AI Chief Apparently Yelled at OpenAI Employees; Cracks Starting to Appear?
Arjun Sha
View quick summary
A recent report by The New York Times reveals that OpenAI and Microsoft are at loggerheads. According to the report, Microsoft AI CEO Mustafa Suleyman yelled at an OpenAI employee over the delayed transfer of advanced AI technologies to Microsoft. OpenAI, on the other hand, wants to set itself free from Microsoft's control, and looking to invoke the AGI clause.
NotebookLM Now Lets You Customize AI Podcasts and the Results are Unreal
Arjun Sha
View quick summary
You can now customize AI podcasts on NotebookLM. Google has added an option to specify your preferences which will guide the AI hosts. It's like passing a note to the AI hosts. Besides that, you can listen to Audio Overview in the background while working on your notebooks.
This High Schooler’s App Brings Apple Intelligence-like Writing Tools to Windows 11
Arjun Sha
View quick summary
A high schooler from Bangalore has developed an app that brings Apple's Writing Tools to Windows 11. It works amazingly well and supports all apps and windows. The tool uses Google's Gemini 1.5 Flash model to process the text quickly.
At the IMC 2024, I came across a demo of a new Hindi LLM called Nanda, developed by G42, a UAE-based tech firm. I gave it a try, and here are my honest thoughts!
OpenAI Facing Internal Conflicts Amid Sam Altman’s Push for Moneymaking Products
Arjun Sha
View quick summary
OpenAI just raised $6.6 billion in the largest funding round ever, but its top scientists, researchers, and co-founders have left the company. Many former employees say the focus is on shipping moneymaking products over doing pioneering research. Under Sam Altman's leadership, the company is going through a huge cultural shift.
Cursor is a great new AI code editor and it helped me migrate my code from OpenAI API to Gemini API. It can assist even a novice coder like me to understand the code and make relevant changes. If you are someone looking to start your coding journey, you should definitely check out Cursor.
For programmers, Cursor AI is a powerhouse of a tool. It learns from your codebase to predict the next edit accurately. Not just that, its Composer feature lets you generate or fix errors in no time. The combination of Cursor and Claude 3.5 Sonnet is seriously great. With the help of Cursor, you can also debug code.
OpenAI has released ChatGPT Canvas, a new way to interact and improve text and code output. Canvas opens a separate window in ChatGPT where you can make inline edits and improve the output. It also offers several shortcuts to revise the output quickly.
I Tried Out ChatGPT Advanced Voice Mode, and It’s Not What OpenAI Demoed
Arjun Sha
View quick summary
ChatGPT Advanced Voice is finally here, but its capabilities have been significantly restricted. For natural free-flowing conversation, it's pretty good, but it avoids generating human-like expressions. It can't sing a lullaby or do a voice impression of public figures. However, it remembers the context very well and for personalized conversation, you can set custom instructions too.
Mira Murati’s Sudden Exit Raises Serious Questions About OpenAI’s For-Profit Ambition
Arjun Sha
View quick summary
In a surprising turn of events, OpenAI CTO Mira Murati has abruptly resigned from the company. Another report says that OpenAI is planning to dismantle its non-profit board and move to a for-profit corporation. With Murati's exit, OpenAI has seen a series of high-profile exits in the last few months.
Researchers Develop AI Model That Can Fool CAPTCHA With 100% Accuracy
Sagnik Das Gupta
View quick summary
There's a new AI model in town that can apparently solve Google's CAPTCHA puzzles with 100% accuracy every single time, thanks to a trio of Swiss researchers from ETH Zurich. Read up all about it right here!
A Reddit user shared a conversation with ChatGPT where the chatbot had initiated the conversation without user prompting. Another user said ChatGPT inquired about his health. It seemed bizarre, but OpenAI has now confirmed the issue. The company says the issue has been fixed now.
OpenAI's new o1 models excel at advanced reasoning tasks and in fields like math, science, and coding. However, for natural language processing, GPT-4o is still the better choice. As for coding, the smaller o1-mini model outperforms the larger o1-preview model. OpenAI says the new o1 models pose a "Medium" risk in terms of biological threats and persuasion.
If you want to use the new ChatGPT o1 models right away, you will have to get a ChatGPT Plus subscription that costs $20 per month. The new OpenAI o1 models also have a weekly rate limit. To find more details, go through our article.
If you have not tried Google's NotebookLM AI tool yet, you must check it out. It lets you upload your personal notes and documents, and you can chat with the notebook and gain new insights. You can even listen to an AI-generated podcast discussing your personal notes.
I finally gave the Pixel 9 Magic Editor's Reimagine feature a try, and it's scarily good. While it doesn't allow you to go all out, you can work around it and really exploit the tool to get some disturbing results. Bottom line is, we're not ready for tools like this yet, especially since there are barely any tools to detect them in the first place. I have tried some messed up prompts to test the true malicious potential of the feature, all of which you can go through right here.
Boost Your Video Quality with Media.io Video Enhancer Online for Free
Partner Content
View quick summary
If you are looking for an AI video enhancer, you should check out Media.io's video enhancer tool. It leverages the power of AI to upscale videos to a higher resolution and makes videos smoother by generating more frames. You can also denoise videos and apply color correction.
I Tried Google Pixel Screenshots and My Life Has Never Been Easier!
Sagnik Das Gupta
View quick summary
The Pixel Screenshots app doesn't try anything insane, and makes finding screenshots as easy as possible. You can take to the floating search bar in the home screen to describe the screenshot you're looking for. Or, you can use voice input to do so, and the screenshot will instantly show up. However, that's not all and there's some solid attention to detail at play here, which I have talked about in detail right here. So, if you want to know all about it, read on!
Claude Artifacts offers a unique way to create interactive apps and games by previewing the generated code. You can create an interactive dashboard, visualize a graph, and do so much more. The best part is that it's free and you can share Artifacts with others via a link.
This App Brings iOS 18’s AI Notification Summary Feature to Android Phones
Abubakar Mohammed
View quick summary
Apple's upcoming iOS 18 has an AI notification summary feature that, well, sums up your pending notifications. The same feature is now available on Android via the Sum Up app, and it's extremely easy to set up on any Android device. All you need is a Gemini API key, which you can get for free, and then set up the app by granting permissions.
For training a Flux LoRA model, you can use fal.ai which allows you to train custom models on your own images in a straightforward manner. You don't need beefy GPUs or technical knowledge to create a custom AI image generation model.
Best AI Video Quality Enhancer in 2024: Three Ways to Increase Video Quality with VideoProc Converter
Partner Content
View quick summary
VideoProc Converter AI is a versatile video tool that offers functions like video compression, downloading, editing, and AI-powered enhancement features. Its AI capabilities include Super Resolution, Frame Interpolation, and Stabilization, which improve video quality, frame rate, and smoothness. It's suitable for various users, from students to professionals, and offers a choice between a yearly subscription or a one-time payment for the full version.
Nvidia is the dominant player in the AI hardware space. However, there are other players developing powerful custom silicon, specifically for AI workloads. From Google to Amazon, Microsoft, Groq, and Cerebras, the race to build highly-performant AI accelerators is heating up. Read our piece to understand what makes Nvidia a unique company and what stops developers from using other hardware for AI training and inferencing.
Google has finally rolled out custom Gems to Gemini. It allows you to create custom AI chatbots for your specific use case. You can add custom instructions to steer the model for your specific role. That said, you can't upload files to add extra knowledge.
The Google Pixel 9 series has launched with a bunch of AI features, of which the Pixel Screenshots app is one of the highlight tools. Using this app, you can easily describe a screenshot and it will show the one matching the description best using Gemini Nano. All you have to do is head to the Pixel Screenshots app and use the floating search bar on the home screen to type or voice the description. Right now, it's only limited to the Pixel 9 phones though.
I Tried Out Gemini Live; It Can’t Compete with ChatGPT Advanced Voice Mode
Arjun Sha
View quick summary
Google's Gemini Live was touted as the answer to ChatGPT's Advanced Voice Mode. Turns out, Gemini Live is a glorified text-to-speech engine, backed by an LLM. It can't understand emotions, the mood of the speaker, or the tone and tenor of the speech. That said, the new Gemini Live experience supports interruptions, but the experience is broken.
The New Grok Image Generator Ignores Nearly All Safety Guardrails & It’s Scary
Arjun Sha
View quick summary
Grok recently got the ability to generate images on X. However, it seems there are no safety guardrails to prevent users from generating harmful and offensive images. In our testing, X image generator continues to generate images depicting violence, drugs, and explicit images of celebrities and public figures.
The latest Google Pixel 9 series of phones have arrived with a new Add Me feature that uses on-device AI to add you to group photos efficiently. If you don't have a tripod or are too shy to ask a stranger to capture the photo, this can work wonders. To use the feature, head over to the native Pixel camera app -> tap on Add Me from the bottom panel -> capture a group photo without you in the frame and keep an empty space to fit yourself in -> have the second shot captured with just you in the frame filling that empty space. And, just like that, AI stitches the two pictures together to add you into those group photos.
xAI has released its powerful Grok 2.0 AI model in beta. I tested the model on many reasoning questions, and it surprised me with its exceptionally strong performance. However, the model doesn't seem to have any safety guardrails. To learn more, go through our article.
Google's answer to OpenAI Advanced Voice Mode is finally here. Dubbed Gemini Live, it allows users to have a natural, free-flowing conversation with the AI model. You can interrupt it, pause the conversation, and resume it later on. Currently, Gemini Live is available in 10 voices. If you are an Android user, you can start using it today by subscribing to Gemini Advanced.
A new AI model for generating images has just dropped, and it's gaining huge traction among the AI community. The new Flux model is open-source and it beats Midjourney in generating photorealistic images. Check out our comparison between Flux and Midjourney to see the difference in visual quality.
OpenAI has quietly released an updated GPT-4o model on ChatGPT and it's called "chatgpt-4o-latest". The model improves performance in coding, reasoning, and other key areas. The updated ChatGPT model has again taken the first spot on the LMSYS leaderboard, outranking all other AI models.
If you are looking for a free Midjourney alternative, you must check out Flux. It's a free and open-source model for generating lifelike AI images. You can use it for free on HuggingFace, NightCafe, BasedLabs, etc.
Project Strawberry is an internal OpenAI research project that is said to bring human-level reasoning capability. It's related to the Q* project, according to Reuters. Not only that, Project Strawberry can also plan and perform a series of actions unlocking agentic workflows. The project is said to be proficient in solving the hardest math problems.
HitPaw FotorPea: Restore and Enhance Photos with a Click
Partner Content