OpenAI's content moderation

PLUS: Even more Google Search AI?


Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • Kneron’s new AI chip 💽

    It’s up to 75% more energy-efficient than existing GPUs.

  • OpenAI uses GPT-4 to moderate content 🤖🚨

    The goal: less involvement from human moderators.

  • More Google Search AI tools 🔎🔥🐜

    AI definitions, coding insights, and SGE while browsing.


Kneron’s new AI chip 💽

Kneron plans to release the KL730 later this year. It’s a more powerful and energy-efficient NPU than the 2021 version — the KL530.

Letters and numbers.

It hopes to directly compete with NVIDIA.

Good luck.

No diggity:

1/ Kneron’s CEO says that its chips are designed specifically for AI.

2/ He said it’s unlike those originally made for processing graphics and forced to work for AI.


3/ It’s up to 75% more energy-efficient than GPUs.

4/ It would allow users to run LLMs entirely offline. This allows more secure data and eliminates the need for a cloud provider.

Our take: Taking on NVIDIA is daunting, to say the least. Kneron didn’t disclose the KL730’s price tag, but if it saves 75% on the energy bill and comes in at a competitive price — things may have gotten interesting.



A Neural Processing Unit is a microprocessor tailored for AI algorithms. It’s similar to a GPU (Graphical Processing Unit), but it primarily operates on predictive models like Artificial Neural Networks and Random Forests.


Your AI Research Assistant

Ever found yourself up at midnight, 20 tabs open, staring at a report due tomorrow that you’ve barely started?

Sounds like you need Upword.

Upword helps you capture notes, summarize information, and answer all your questions. It’s got all the power of GPT-4, personalized to your knowledge.

It does the hard work for you. Like an intern you barely have to pay… or feed.

Get your research done 10X faster using the GPT-powered research tool built to help you understand the bottom line of what you need to know, easily and fast.


GPT-4, the content mod 🤖🚨

OpenAI has revealed that it uses GPT-4 for content moderation. The result: streamlined content moderation/labeling and faster policy updates.

AI inception?

The goal is to reduce the dependence on human moderators and to reliably scale content moderation.

How it works (simplified):

1/ Human experts label queries as safe or unsafe.

2/ GPT-4 refines its responses to match these labels.

3/ A compact version of this refined model handles large-scale content moderation.

The tea:

1/ OpenAI says GPT-4 reduces policy change times from “months to hours.”

2/ It creates more consistency, a faster feedback loop, and a reduced mental burden on humans.

3/ OpenAI says it’s “more reliable“ than Anthropic’s Constitutional AI.

Our take: seeing AI take over content moderation is probably a move for the better. We’ve seen this before, but doing so on a wider scale will mean a huge reduction of psychological burden on human mods.

And considering how closely content moderation is tied to advertising dollars, this change may — or may not — cause some big waves. It all depends on how well the AI works.


Google Search’s AI features 🔎🔥🐜

AI-powered definitions, coding info in overlays, and SGE while browsing.

Plus, more censorship?

The juice:

1/ Hover over unknown words for complete definitions.

2/ Code in search results becomes color-coded with syntax highlighting.

3/ “SGE while browsing” summarizes articles for you — (non-paywall ones).

Our take: Google’s gone ham on search. In August alone, they implemented:

And now, all this as well.

Hot dog!



Think Pieces

AI recreates a Pink Floyd song from brain scans. How? By monitoring the brain activity of test subjects, listening to “Another Brick in the Wall.”

ChatGPT vs Stack Overflow. A Purdue University study shows that ChatGPT wins, though the number of errors is “alarming.”

How AI creates misinformation. How it does it and what it means for politics and society.

Startup News

VoiceFlow raises $15 million. The platform allows teams to collaborate and create AI agents to automate voice and chat — primarily for uses cases like customer service.

DeckMatch raises $1.1 million. The startup is automating investments with AI.

WhatsApp tests AI stickers. It currently experiments with them in Android Beta.


Moderation in GPT-4. OpenAI’s API docs on content moderation.

SpeechX — a speech-to-text model that preserves background noise during edits and leverages transcriptions for enhancements.

Platypus — a paper on the dataset of fine-tuned/merged LLMs that’s number 1 on the Open LLM Leaderboard.

VisIT-Bench — a paper on a benchmark for real-world use of vision-language models. It stands for Visual Instruction Benchmark.


UpWord [Sponsored] — your own personal ChatGPT-powered research assistant. It captures notes, summarizes information, and answers all your questions.

Function AI — a tool to run AI prediction functions.

ClearMind — AI-powered therapy tool to measure mental health.

Grants Finder — search for AI grants and credits. $8 million found so far.

Looks cool

Autonomous Agents Hackathon — Hackathon hosted by LabLab AI with $10k in prizes for projects built using autonomous agents.


CEO of Stability AI doing his fair share of showboating.

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.


🏥 Healthcare AI News — 5 minutes or less is all it takes to elevate your Healthcare knowledge with this expert-curated weekly AI newsletter. Stay informed and stay ahead.

👨 The Average Joe — Market insights, trends and analysis to help you become a better investor. We like their easy-to-read articles that cut right to the meaty bits.

💌 Marketing Letter — The newsletter keeping 30k+ marketers in the loop. Read by marketers who work at LinkedIn, Techcrunch, and Disney.


Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.