Daily Digest: Data shops are open for business.

Want to get in front of 100k AI enthusiasts? Work with us here

Hello folks, here’s what we have today;

PICKS

1. Reddit is getting serious about AI and its upcoming IPO. And looks like they've signed a deal for $60M every year to let an AI company (name’s hush-hush) use their content for model training. But that number feels small.🍿Here’s why (also below)

2. OpenAI’s fundraising deal valuing it at $80B seems to have been completed. This deal has been in limbo since Sam Altman’s firing in late November. No official comments though.

3. Groq (not the Elon one) is serving AI models at insanely fast speeds. The claim is that using their new LPUs (stands for Language Programming Units) is faster than GPUs for serving LLMs. Matt built a demo with Groq’s instant answers, check it out.

TOP TOOLS

• AI Sound Effects by ElevenLabs - Generate realistic sound effects. Coming soon.

• Mistral Next - Mistral’s new GPT-4 class model is available on LMsys for testing. Click “direct chat” in the top menu to use it.

• DomoAI Video to Video 2.0 - Animate your videos, now in Ghibli style too.

• Custom Voice by Clubhouse - Text a friend, but they get a voice message, in your voice.

• Repeto - Learn smarter with AI by your side.

• Emojifyer - Emojify yourself.

View more →

NEWS

• How Sora works and what it means.

• Anthropic will update their prompt shield to prevent election misinformation.

• Masayoshi Son seeks to build a $100B AI chip venture.

• Raising children on the eve of AI.

• Microsoft plans to invest €3.2B in Germany for its AI push.

• Women In AI: Irene Solaiman, head of global policy at Hugging Face.

• Microchip maker Graphcore could sell for $500M+.

• AI hiring tools may be filtering out the best job applicants.

• Building a RAG-based digital restaurant menu.

• Khan Academy’s chatbot teaches students but is bad at math.

View more →

QUICK BITES

Reddit is getting serious about AI and its upcoming IPO. And looks like they've signed a deal to let an AI company (name’s hush-hush) use their content for model training.

What is going on here?

Reddit is licensing its content for training AI models, to get a cash boost before a potential IPO.

What does this mean?

Bloomberg reports that it's a $60 million annual contract with a big (but unnamed) AI player. Reddit's basically saying, "Hey AI folks, wanna train your models on the wild conversations and random stuff our users post? It'll cost you."

Last year Reddit tightened up how people could access its data (strict rate limits, pricy API plans, etc.). One key reason was Reddit wasn’t getting much in return for its data that supposedly powered these AI models. And now bang on, Reddit got that bag.

Big question is: did Reddit hand over ALL their training rights to one player?

I keep hearing: data is the new oil. $60M looks like gas station money, not oil well money. It’s better than nothing, but last year, Reddit raked in over $800M. That’s less than a 10% annual bump. It could make sense if the deals are not exclusive, and Reddit partners with a dozen more foundational model providers.

Why should I care?

Parent companies of Youtube, Facebook and Twitter all are creating their own models. But for other social media companies who aren’t, the future’s clear: Restrict training on the content mine you have, then license that content for a payday. Fair for business, but I wonder how users will react to this.

On the other hand, conversational chat data would impact how AI chatbots interact. No one wants them to keep blabbering, most of us want that snappy, chat-with-a-friend feeling. So, the POV depends on how often you are using Reddit and AI tools.

Share this story

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

• All 10k+ links we’ve covered, easily filterable (1 referral)

• 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)