🍭 Claude Gets Buff. Benchmarks Get Bent. 🏋️

Good morning. Claude’s been poking around my Google Drive like he’s my new product manager. And honestly? He might be better at it.

What’s Cookin’:

• Anthropic just gave Claude a toolbox

• Google Ads showing up…

• AI Arena accused of leaderboard rigging

• 5 AI tools, from mental health bots to research unlocks

  
  Your annoying friend fwd this to you? → (Sub here for free)

Anthropic
🤖 Claude gets a power-up: Anthropic goes full app mode

Claude just got a major upgrade.

The Bite: Anthropic launched Integrations and Advanced Research, two new features aimed at helping Claude play nicer with tools, browse better, and compete harder with ChatGPT and Gemini.

Snacks: 

• Integrations: Claude can now connect to apps like Atlassian, Zapier, Intercom, Square, and PayPal

• Advanced Research: Crawl hundreds of internal/external sources, then deliver deep-dive reports in 5–45 minutes

• Claude will break down your request into parts and give full citations

• Claude Code also gets higher usage limits for coders

Why it bites: Claude just went from smart friend to “do-my-work-for-me” friend.

If Zapier and Confluence had a baby... Claude would be its helpful, overachieving cousin.

AI Benchmarks
⚔️ ‘LM Arena’ is helping labs juice up benchmarks?

The Bite: A new study from Stanford & MIT says LM Arena quietly let major labs like Meta, OpenAI, and Google rig the game… are you surprised though?

Snacks:

• Select labs got to privately test dozens of models without publishing bad scores

• Meta tested 27 Llama variants before only releasing the top performer

• More battles = more feedback = unfair edge on Chatbot Arena's leaderboard

• LM Arena says the study is flawed, but confirmed many of the claims

• Suggested fixes include caps on private testing and equal sampling rates

Why it bites: If AI models are graded on a curve, but some students saw the exam early... we got bigger problems than just model bias or glazing.

Building LLMs to pass benchmarks instead of overall model quality is a human problem… not an AI problem. Jus sayin’ 🤷‍♂️

Source: TechCrunch

🍭 Steal this prompt
Make your brand look good enough to eat

Ever wanted your logo, icon, or product photo to look like a gummy candy? 

Me neither, but now you can do it before you finish your coffee. 👇

Here’s the workflow:

1. Hit this link

2. Copy + paste the prompt into ChatGPT 4o

3. Upload your logo, icon, product photo, or literally anything!

4. Chill and let the candy factory cook. Don’t forget to brush your teeth after

What makes this prompt extra AF:

• Preserves your shape and colors like a sugar-coated clone

• Gives anything — logos, icons, objects — a glossy, chewy candy finish

• Cranks up the realism: jelly-light refraction, sticky shadows, edible-level detail

  
  🍭 Try the prompt on Snack Prompt 🍭

Do More. Spend less on SaaS.

Launch, grow, and scale your company faster with Notion.

Thousands of startups rely on Notion to move quickly, stay aligned, and replace multiple tools. Whether you're building a wiki, managing projects, or writing documentation, Notion is your all-in-one workspace.

Get up to 6 months of the new Plus plan + unlimited AI, for free!

To redeem your Notion for Startups offer, simply visit the Notion for Startups page and apply.

Get the offer

Test Your Skill
Snack Quiz: Choose The Real Image 🔍

Google
 💸 AdSense creeps into chatbots

The Bite: Google is now placing ads inside AI chatbot convos starting with partners like iAsk and Liner.

They said AI would kill search. Google said: “bet,” and started selling ad slots in your chatbot.”

The Snack:

• Called AdSense for Search, it's now available for chat-based apps

• Targets users mid-convo, just like how sponsored answers show up in Google Search

• Strategy: monetize the shift from web search to chatbot search

• This isn’t no-code or drag-and-drop: it’s real dev work, done by Copilot and AI tools (we use Cursor but hey…tomato/potato am I right?)

Why it bites: Imagine your chatbot telling you to "consult a doctor"... and then linking you to sponsored Theraguns.

Welcome to the AI ad economy… Intern finally spoke up; “i got an idea for a new kind of ad-blocker”

Tool Box
🧰 Trending AI Tools

📊 Trend Hunter: Ask it for product ideas or trend reports, and it’ll hit you back with AI-fueled business foresight. Your R&D team’s new favorite coworker.

🕵️‍♂️ v0 by Vercel: Describe a UI, get working React code. Prototyping just got vibe-coded.

⚡ Woebot & Wysa: Mental health chatbots. (But remember what we said yesterday about sycophantic LLMs...)

🎬 OpenAI Deep Research: Finally open to free users. Hit the toggle. Explore like a scholar.

Pro Tip: Lots of context and clearly state your motivation, goal, and desired outcome of the research. (it’ll still ask you a bunch of questions, but they’ll be more refined if you lead with a ton of context)

🔎 Flora: A visual playground for AI workflows.

Drag, drop, and remix models like it’s Figma for generative media.

Everything Else
You need to know 📈 

🎨 Gemini gets native image editing.

Google's Gemini now lets you edit uploaded photos and AI-generated images natively.

Background swap? Object replacement? You got it.

📲 World expands ID tech via Tinder + Visa

The eyeball-scanning Orb company teams up with Tinder Japan, Visa, Kalshi, and more to bring biometric ID to the masses.

ngl… “eyeball-scanning…company” sounds pretty creepy out of context…

🤖 Amazon launches Nova Premier

Its most powerful AI yet. Multimodal. Million-token context window. Slightly worse than Gemini in coding, but wins in retrieval.

🧰 Crowd-trained LLMs are here (our personal fav of the day)

Flower AI + Vana quietly dropped my favorite tool rec for today: a crowd-trained AI model…no data center required.

It’s called Collective-1, and it’s built using decentralized GPUs and private data from Reddit, Telegram, and X. Think of it like BitTorrent meets GPT.

Referral Program
Invite Friends & Earn Some Freebies 🔍

🦍 100 Men vs. 1 Gorilla 🦍

The most important debate of 2025.

No, this isn’t a new MrBeast challenge (yet). TikTok, Twitter, and Reddit are currently locked in a primal battle of brains vs. brawn: Could 100 unarmed men beat 1 silverback gorilla?