Sign up102 | Advertise53 | Ben’s Bites News44
Daily Digest #275
Hello folks, here’s what we have today;
AI Halloween - Andrew Ng fears the sensationalist claims about human extinction436 are doing more harm than good. Andrew evaluates the legitimate risks while calling out fear-mongering and financial incentives behind it.🍿Our Summary291 (also below)
Phind Model beats GPT-4 at coding945† - Phind's new code generation model beats GPT-4 at coding tasks while running 5x faster at 100 tokens/second. Beyond the benchmarks, Phind claims real users find the Phind Model as helpful as GPT-4 for answering technical coding questions.🍿Our Summary352 (also below)
The executive order aftermath:
What the executive order means for openness in AI.471
Regulating AI by executive order is the real AI risk.367
A free tool356† for DevOps Prompt Engineers - from Shoreline.io170†*, a pioneer in AI-driven production operations.
StockMusic648 - Generate upto 10 minutes of copyright-free music.
Local AI stack470 - Local-only AI apps starter kit to build AI apps that cost $0 to run and test.
LangChain Templates523 - A collection of production-ready reference architectures.
Nintee555 - Break bad habits and pursue goals with an AI coach.
Trickle502 - Transform screenshots into searchable treasures with AI.
Robin AI’s Contract Copilot302 is now available as Microsoft Word plugin for lawyers to save time and money.
Gems483 - Get ready-to-use answers from all your knowledge.
View more →243
*sponsored
Fixie156 - The conversational AI app platform.
Elicit124 - Speeding up research paper analysis.
contextSDK78 - Real-time user context to improve engagement.
Rewind176 - Your AI meeting note taker that's not a bot.
LlamaIndex113 - Helping developers build applications using LLMs and their data.
Two new memos from our friends at Contrary Research: Midjourney267 and Neuralink196.
Distil-Whisper from HuggingFace172 - 6x faster while sacrificing only 1% on short-form eval and beating Whisper on long-form eval.
Healthcare startup Olive AI sells off business units, winds down279 after raising $900M and a $4B valuation.
SAP joins Stanford HAI87 corporate affiliate program.
Who’s going to the UK’s AI Safety Summit?64 Cambridge, MIT, Russia and Apple are not going. Official list here80.
New nonprofit backed by crypto billionaire scores AI chips worth $500M.149
How to deploy your model on Replicate124 for inference.
$95 AMD CPU becomes 16GB GPU234 to run AI software.
Working with AI: Two paths to prompting.359†
In the past few days, the heat between legends of the AI realm has been increasing. The battle is about the risks of AI436. Andrew Ng, the co-founder of Google Brain and Coursera believes that big tech is pushing the human extinction by AI theme for their benefit and other researchers are adding fuel to the fire.
What is going on here?
Andrew believes while AI risks exist, fear-mongering causes overblown anxiety. Baseless warnings of AI-driven human extinction could impede AI progress that benefits humanity.

What does this mean?
Sincerity exists, but so do individuals/organizations who have strong financial incentives by overflowing the human extinction and superintelligent AI stories. Here’s a quick list of some legitimate fears and solutions in progress:
1) Hallucination that leads to misinformation, but developers are improving models and adding filters.
2) Easy access to tools for cybercrime: deepfakes, voice clones, malicious bots etc. Governments and companies are diverting resources to red teaming.
3) Copyright disputes for training data breach trust, but companies are finding legal ways to access data.
4) Fears that AI will eliminate human jobs are exaggerated since AI tends to transform jobs rather than replace them outright.
5) AI startup hype may lead to a correction, but the usefulness of generative AI means it's here to stay.
Why should I care?
While the researchers are debating pro and cons like they are their dorm room, take a step away. Think about who has what incentives. Dig deep to evaluate if the claims hold on basic interrogation.
If you’re a student, a builder, or a developer, focus on finding the legitimate fears and facing them head-on.
Scary Goofball AI - Fun little games for Halloween made by GPT-4, DallE-3 and Midjourney.
Angry Pumpkins127 - Pumpkins are better than Birds, right? (process here78)
BooseGumps108 - Tribute to the choose your own adventure series.
Phind has released a new model for code generation945† that matches or exceeds GPT-4 in quality while being much faster. Beyond the benchmarks, Phind claims real users find the Phind Model as helpful as GPT-4 for answering technical coding questions. Phind is a search engine + pair programmer using AI to help developers.
What is going on here?
Phind's new code generation model beats GPT-4 at coding tasks while running 5x faster at 100 tokens/second.

What does this mean?
The Phind Model V7 achieves a 74.7% accuracy on the HumanEval benchmark. The model is built on top of Phind’s earlier finetunes of CodeLlama-34B. After additional fine-tuning on 70B tokens of code, the Phind Model V7 now provides GPT-4 level responses in 10 seconds rather than 50. This speed boost comes from optimizing the model to run on NVIDIA H100s using the TensorRT-LLM library.
The Phind Model also supports up to 16k context, allowing for more detailed coding questions. Currently, 12k tokens are enabled for input on the website with 4k reserved for search results.
Why should I care?
Phind agrees there are still some rough edges, for example on certain challenging questions, the Phind Model might take more generations to get to the right answer than GPT-4.
Even with those limitations, getting GPT-4 level answers in 5x less time is impressive. Access to a larger context also allows including more code samples and details in your questions.
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)
Daily Digest: What AI legends fear
PLUS: new coding champion is fast
3,633 readers clicked at least once. 43 links, 43 with clicks. Heat is relative to the most clicked link in this issue.
Beehiiv · digest