BB Digest: Is Copilot under attack?

PLUS: Github's free Devin, Reddit vs Google

January 14, 2025

Digest #523 → Subscribe | Upgrade to Pro

Hey folks, we* are COOKING in 2025 already. We now have a Slack bot @AskBB to recommend AI tutorials to our members. See the smartass in action.

* = Keshav is cooking

TLDR; inside today’s newsletter

  • Cohere’s AI workspace, new coding champs and Reddit’s AI move.

  • ICYMI: new tutorials and blogs for you

  • snack n skill - a bite-sized lesson to analyze Reddit trends

  • 7 new AI tools for video editing and local AI

  • How o1 differs from other models + my tiny experiment with it

  • 6 interesting posts including UK's AI plan

  • to-do's

Alright, let’s get to it!

  • North, Cohere's new AI workspace for enterprises, combines LLMs and automation for air-gapped environments. It claims that North (in early access now) outperforms Microsoft Copilot and Google's Vertex AI across finance, HR, and IT benchmarks. Cohere’s an underrated company that pivoted its model development strategy from “raw intelligence” to workplace utility—worth looking into.

  • Github’s coding agent - Copilot Workspace is now open to everyone in technical preview. No more waitlist and it’s free for now (demo). Let’s see what the final pricing will be—Devin’s plans start at $500/month.

  • Mistral has a new coding model - Codestral 2501. It is now #1 on Copilot Arena by LMSYS (together with DeepSeek and Claude). It comes with a longer context window (256k), low latency outputs and is available via continue.dev in major IDEs.

  • Reddit is introducing Answers in early access - You can ask AI to find answers to your questions from discussions on Reddit. We covered it earlier but it got buried under December’s craziness. This one will be interesting to watch for Google—it’s under attack from all sides.

  • Supercharge your coding with AIDE—the AI-native VSCode fork. Enjoy the SOTA agentic framework (SWEbench-Verified) plus unlimited Sonnet 3.5 & o1-mini for $20/month. Code smarter & faster with agents in your editor. Try it today at aide.dev*
    *sponsored

👀 ICYMI: New stuff we’ve launched

🍽️ Snack ‘n’ skill

Learn how to analyze Reddit trends using AI

Ever wish you could spot emerging trends and discussions across Reddit without spending hours scrolling? You can analyze thousands of Reddit conversations using GummySearch's AI tools to uncover patterns, themes, and growing topics in your niche.

Step 1. Create your custom Reddit audience

Build a targeted data pool by selecting keywords and subreddits that matter to your business. Start with broad terms like "Artificial Intelligence" and add niche keywords to expand your subreddit selection. GummySearch automatically finds relevant communities.

Step 2. Mine insights from Reddit discussions

  • Use keyword search to find specific discussions and get AI-generated summaries of entire conversation threads

  • Explore "Hot Discussions" to see what's trending in your audience over the past week

  • Check the "Solution Requests" theme to understand what problems your audience is trying to solve

  • Click "Ask" to question the AI about specific patterns or trends in any theme

  • Track growing subreddits to spot emerging communities in your niche

  • Save important searches with the bell icon for continuous monitoring

Tip: Don't limit yourself to a single keyword when building audiences. Add related terms to discover niche subreddits you might have missed.

This walkthrough is a condensed version of this tutorial. Check it out for more details on how to use Reddit for social listening.

⚙️ Top new tools

  • Trellis AI - Automate your PDF workflows at scale.

  • Captions, the AI video editing app now has a free plan.

  • Seam - Agents for sales teams to find lookalike companies, contacts, and research prospects.

  • Aidy - Around the clock research on issues that drive your organization.

  • Retape - Create personalized demo videos with AI.

  • Fullmoon - The simplest way to use local LLMs like Llama on your iPhone. (These are baby models—not too smart, but give Siri a run for its money)

  • GenUI - AI chat that transforms itself as you talk. (read more)

🗣️ Everyone’s talking about… how o1 differs from other AI models

OpenAI released a new line of reasoning models o1-mini and o1-preview a few months ago and last month they announced a $200/month plan for ChatGPT with a Pro mode to use o1 to max limits. But there’s still confusion about how it differed from “regular” models like GPT-4o, Claude 3.5 Sonnet and Gemini 1.5 Pro.

Recently, Ben Hylak wrote a post describing the difference between prompting regular AI models vs reasoning models like OpenAI o1. o1 isn't meant to be used like a chat model. Instead of back-and-forth conversations, it works best when you provide extensive context upfront and let it reason through the problem in one shot.

Here’s a tiny example from the other day:

I was going through a code example that called itself “Auto-RAG”. I wasn’t sure so I put the AutoRAG paper’s abstract and the entire code in a ChatGPT message and asked GPT-4o: Does this code qualify to be called Auto Rag?

GPT-4o (and Gemini too), incorrectly, agreed that it was. That query was perfectly set for o1 too—enough context + a single goal. o1 spotted the key difference and explained why the answer should be no.

You can use o1 for similar “decision-making” queries where it has to reason over a given definition or criteria. For example:

Does this contract cover what we talked about in the email? Does it misinterpret/remove/add anything we talked about? <add email text> + <add contract text>

Let me know how are you using the o1 models.

 📜 Interesting posts

📌 To-dos

  1. Let us know if you have any tutorials or courses you want us to create—we’re always open to ideas.

That wraps things up for today! See you again on Thursday. 👋

Ben

Enjoy this newsletter? Please forward to a friend.

Want more education on AI? Become a Ben’s Bites member and access our courses & tutorials.

Want to advertise in this newsletter? Click here.

Reply

or to participate.