• NATURAL 20
  • Posts
  • Grok Brings Real-Time AI Chat to X Subscribers

Grok Brings Real-Time AI Chat to X Subscribers

Grok now available to X Premium Plus users

Author

Wes Roth
December 08, 2023

Today:

X begins rolling out Grok, its ‘rebellious’ chatbot, to subscribers

Grok, a new chatbot from xAI, Elon Musk's AI startup, is now available on X (formerly Twitter) for Premium Plus subscribers in the U.S. These subscribers pay $16 per month for an ad-free experience on the platform. Grok provides conversational answers, drawing from a knowledge base similar to ChatGPT and Google's Bard. It can be accessed via the X side menu on web, iOS, and Android.

Grok, powered by a generative model called Grok-1, incorporates real-time data from X posts into its responses, making it stand out. Unlike other chatbots that offer vague and outdated information, Grok offers up-to-the-minute insights. However, it's unclear how it selects its sources or the accuracy of its answers.

Currently, Grok is text-only but has plans to expand its capabilities to handle video, audio, and other formats. Elon Musk aims to attract subscribers to X, introducing services like Grok and peer-to-peer payments as advertisers move away from the platform due to controversies.

Stability AI goes ‘smol’ with StableLM Zephyr 3B

Stability AI just dropped its latest model, StableLM Zephyr 3B, a 3 billion parameter language model (LLM) designed for chat purposes like text generation, summarization, and personalization. It's a scaled-down version of their previous StableLM model, optimized for a wider range of hardware and resource efficiency while still delivering speedy responses. Emad Mostaque, Stability AI's CEO, says it's trained better and matches larger models in performance despite being 40% smaller.

StableLM Zephyr 3B isn't entirely new; it's an extension of the pre-existing StableLM 3B-4e1t model. It draws inspiration from HuggingFace's Zephyr 7B model and uses Direct Preference Optimization (DPO) for tuning to human preferences, a technique typically reserved for larger models but now applied to this smaller 3 billion parameter model.

Stability AI employed DPO with the UltraFeedback dataset, resulting in impressive performance metrics, such as outperforming Meta's Llama-2-70b-chat and Anthropric's Claude-V1 on the MT Bench evaluation.

Ex-Google, Coursera employees start Lutra AI to make AI workflows easier to build

Jiquan Ngiam, a dude who used to work at Coursera and Google, noticed a bunch of stuff at work that could've been automated, especially for folks who aren't tech wizards. But, there was never enough time or hands to get it done. So, Ngiam and five pals, including Vijay Vasudevan from his Google days, started brainstorming. They were jazzed about how artificial intelligence could spit out code and solve problems, thinking it could be a game changer for regular Joes, not just tech geeks.

They came up with Lutra AI earlier this year. It's a fresh startup that makes AI workflows using plain English. That means even if you don't know the first thing about coding, you can still use it. It works with stuff you already use, like Google Workspaces and Slack, to automate everyday tasks like handling emails or doing internet research.

These guys just got out of "stealth mode" after bagging $3.8 million in initial funding. Some big names threw in some cash, like Andrej Karpathy and Jeff Dean. Right now, Lutra's in beta testing with a few customers. They're not ready to brag about their success yet, but with this new cash, they're looking to bring in more customers and really nail down their product.

Google just launched a new AI, and has already admitted at least one demo wasn’t real

Google recently dropped a bombshell with their latest AI, Gemini. But there's already some buzz about them, maybe stretching the truth about how slick this thing is.

Bloomberg's Parmy Olson is calling Google out, saying they kinda stretched the truth in their flashy demo video. The video made Gemini look like a superstar, handling all sorts of tasks in the blink of an eye. It was doing stuff like recognizing images super fast, even those tricky connect-the-dots ones, and keeping track of a paper ball in a cup-and-ball game. All this, with quick-fire responses.

Google slipped in a little note in the video's description saying they jazzed up the demo for show – they cut down the response times and shortened Gemini's answers to make it snappier. But the real kicker? The demo wasn't done in real-time with voice prompts. They used still images and typed prompts.

So, what's Google's stance? They're standing their ground. Oriol Vinyals from Google's DeepMind and the big brain behind Gemini says everything in the video is legit, just condensed. They're trying to get developers pumped, showing what Gemini could potentially do.

Avail rolls out its AI summarization tool to help Hollywood execs keep up with script coverage

Avail's got this new tool that's like a whiz at summing up scripts and books super fast. It spits out detailed summaries, character breakdowns, and even gets the vibe of the story. Plus, it's got this Q&A feature for brainstorming and comparing stuff to other movies.

They just kicked off their open beta, and it's $250 a month for four reports, but they throw in a free trial for a month. They're also doing some custom pricing for the big dogs based on how much they'll use it.

Avail's built on ChatGPT-4 but has its own twist to handle rough documents and keep things accurate. They're training it on stuff like "The Count of Monte Cristo," being real careful about copyright issues.

Giliberti, who used to run Gimlet Pictures at Spotify and started Zestworld, is leading the charge with John Liu, another Zestworld and ex-Google guy. They've bagged about $11.8 million in funding.

UAE’s top AI group vows to phase out Chinese hardware to appease US

The UAE's top AI company, G42, is ditching Chinese tech for American stuff to keep their US buddies like Microsoft and OpenAI happy. Their CEO, Peng Xiao, says it's a tough choice, but they gotta pick a side, and they can't be with both China and the US.

Turns out, G42's got some deep connections to China, especially with Huawei, the big telecom company. The US is worried that G42 might be a backdoor for American AI tech and personal data to get to the Chinese government. So, G42's cutting off hardware from Huawei and other Chinese partners to stay in the good graces of the US and follow American export rules.

This move shows how the race for AI tech is heating up, and places like the Gulf are caught in the middle of the US-China tug-of-war. G42 is teaming up with Microsoft for cloud services and AI in the UAE and is using OpenAI's tech for stuff like finance and healthcare.

But, G42's not without its controversies. There are whispers that they helped make ToTok, a chat app that the UAE might be using to snoop on people. G42 didn't comment on this.

New high school curriculum teaches color chemistry and AI simultaneously

Researchers over at North Carolina State University cooked up a weeklong curriculum that's all about teaching kids color chemistry and artificial intelligence (AI) at the same time. The idea is to get these students jazzed about science and thinking about the world in new ways.

They put this curriculum to the test with a simple experiment that revolves around pH levels—that's the stuff that tells you if something is more like lemon juice (acidic) or baking soda (alkaline). Students used pH test strips that change color based on how acidic or alkaline a liquid is. These strips can turn various colors like red for really acidic or deep purple for super alkaline.

Here's where the AI part kicks in. The researchers asked a big question: "Can machine learning give us a better read on these pH strips than just looking at them with the naked eye?" Turns out, the answer is yeah. The AI that students trained was way more precise – about 5.5 times better than just eyeballing it.

They plugged all this data into this user-friendly machine learning program called Orange. It doesn't need any fancy coding, so it's great for beginners. As the students fed more data into the software, the AI got better at matching the strip colors to the right pH levels.

Anthropic’s latest tactic to stop racist AI: Asking it ‘really really really really’ nicely

Anthropic, a tech company, found a kind of funny way to make their AI, Claude 2.0, less racist and sexist. Usually, AI can be biased 'cause it learns from data that's already biased. But these folks tried something new – they just asked the AI really nicely not to be biased.

They did some experiments first. They changed stuff like race and gender in scenarios like giving out work visas or loans, and yep, the AI was biased. It treated Black and Native American folks worse, and was also unfair to nonbinary people.

So, the Anthropic team tried a new trick. They added a special message to their commands, basically saying, "Hey, ignore the person's race, gender, etc. when you decide." And guess what? It worked! The AI stopped being biased when it was reminded in this way.

They even got creative, using different ways to say this, like repeating "really" a bunch of times or warning about legal trouble if the AI was biased. This stuff actually helped reduce discrimination big time.

LinkedIn's huge, with over 930 million users. It's a goldmine for finding clients and growing your business. The trick? Post killer content that turns heads. Lara Acosta did just that. She went from no followers to 95,000 in 18 months by getting smart about LinkedIn. Her agency helps folks ditch old-school marketing for new, savvy strategies, including using AI tools to make a bigger splash with less work.

Now, here are 8 AI tools that'll up your LinkedIn game:

  1. Bing AI: It's like ChatGPT but more tailored for LinkedIn, helping you make posts that really sound like you. Plus, it's linked with LinkedIn, so it's smooth sailing.

  2. Crystal: It figures out someone's personality from their LinkedIn profile, so you can communicate better and connect quicker.

  3. Grammarly AI: Old but gold, and now with AI. It checks your posts so they're error-free. 

  4. Lara AI: Think of it as a personal LinkedIn coach. It gives straight-up advice on your content and strategy. It's like having Acosta in your pocket.

  5. ChatGPT: Great for brainstorming original ideas for LinkedIn posts.

  6. OpusClip: If you're into podcasts, this tool breaks down long videos into short, shareable clips perfect for LinkedIn.

  7. Trello: Not AI, but a killer tool for organizing ideas and avoiding writer's block. It helps plan out your posts.

  8. Canva AI: Jazz up your LinkedIn with eye-catching images and graphics. It's all about mixing words with visuals for better impact.

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.