Nibble #16
Claude 2, the ChatGPT killer (another one!), Bard Updates, Elon's X.ai, StabilityAI's Shaka Laka Boom Boom, ISRO's Chandrayan, AI discovers anti-aging, GPT-4 leaked
🙏 Namaste & welcome to Nibble #24
Apologies for the delay in today’s edition. Nibblers were AFK for a lot of time this week.
If you are new to this place, don’t forget to Subscribe.
What’s happening 📰
🆕 Anthropic released Claude 2, a new model with improved performance, and longer responses, and can be accessed via API. Also, a new public-facing beta website, claude.ai. (
ChatGPT blocked at your workplaceChatGPT blocked ya? talk to Claude 😉) [Video]Can summarize up to 75,000 words, a massive leap compared to ChatGPT's 3,000-word limit
Knowledge cutoff is early 2023, unlike ChatGPT, which is in 2021
Outperforms in GPT4 in exams & has a code interpreter already.
It is good with files & can summarize a doc.
Cheaper than GPT-4
🆕 Google’s Bard June Updates - Bard drops their first major update after I/O. It now:
Supports more languages such as Arabic, Chinese, German, Hindi, and Spanish
Speak out responses
Modify the response to be more simple, long, short, professional, or casual
Pin and rename convos and share them via the link
Export code to ReplIt along with Colab
Aaanndddddd finallllyyyyy use images in prompts (This is even more exciting because they shipped it before OpenAI)
BTW, did you notice that lately a lot of companies have been naming their updates as <Month> Update? Also, Monthly Nibble updates might start hitting your inbox, once we find our seed investors 😉
🗺️ Elon has prepped the ship to explore & understand the complete universe with his new startup X.ai (Who’s copying Zuck’s MetaAI now, huh? JK.)
🤝 OpenAI & Shutterstock deals get an extension of 6 years, to provide better Generative AI Image toolings. (it’s better to legally bind with these image providers early in the game, rather than getting sued by them later)
🤝 Talking about OpenAI starting to pay up for the data, Open AI & Associated Press signed a deal for Open AI to get the license to AP’s text archive, and AP gets access to LLMs and cool stuff that OpenAI builds.
🥹 Bruh Moment Report - Cyrus Hodes, co-founder of Stability AI, is taking legal action against the firm and its CEO Emad. He alleges that Mostaque fooled him into giving up his 15% ownership for only $100. Three months later, it received a hefty $101 million investment with a post-money valuation of $1B - resulting in the potential worth of half a billion dollars for Hodes' shares! (Emad has quite a list of people suing him, now!)
🤬 In this week’s dose on people suing techies, Sarah Silverman (author) sues OpenAI, Meta over the use of her book to train AI. (Did she consider that a book review might be the culprit here? Also, a good contender for the Machine Unlearning Challenge)
🦆 Google launched NotebookLM and missed a once-in-a-lifetime opportunity to (aptly) call it ‘Doc Doc Go’. Anyways, Google being Google, it is trying to expand its LLM integrations into the rest of its product offerings and this is an experiment to slap a chatbot to your documents (RIP startups #KilledByGoogle again).
✏️ Stability AI just released 'Stable Doodle' [Blog Post] which lets anyone with
terriblebasic doodling skills turn those silly lines into pieces of art. We are officially in the Shaka Laka Boom Boom age!
Masterpiece by Nibbler P. Now we all know how that childhood scenery really looked. 🪙 TCS Bets Big on Azure Open AI: Plans to Train 25,000 Engineers; Launches New Generative AI Offering for Clients.
🤖 China's Alibaba and Huawei add products to AI frenzy - Alibaba Cloud launched ChatGPT-like text generator Tongyi Qianwen in April & now Pangu 3.0 will mostly serve industrial usages (no more child play like US folks maybe!)
🙈 Twitter will reportedly let you hide your “liked tweets” soon - What? Yes! feel free to like the tweets that you like (all hail sauce tweets!), but be ashamed to let people know. Keep your likes & interest private (just you, Elon & advertisers will know now.)
⛙ GitHub merge queue is generally available - Now you can merge quickly in the busy branches without worrying about the incompatible changes in the PRs ahead of you! The Merge Queues will pull the changes ahead of branches in the queue and run the CI checks on it. So, that you don’t realize that prod is burning when it is actually burning, but you realize it beforehand. (it allows you to look proactive instead of being reactive).
🚀 Chandrayaan - 🇮🇳 Last but not the least, ISRO launched Chandryaan 3, India’s humanless lunar mission. It is expected to do a soft landing on Aug 23rd, 5:47 pm. The budget for this complete mission is ₹615 crore. (which btw is less than a lot of
shittymovies.)
What brings us to awe 🤪
💊 AI might have found the anti-aging drug that 26-year-old developers are looking for (just kidding, we know tech billionaires have an obsession with living longer too, despite some hate)
TL;DR => The class of drugs that prevents aging is called senolytics drugs, and they work by killing zombie cells, which are useless and can’t replicate. So how it works is we humans, collect molecules and test each one to see if they are senolytics, till now we were able to find 80 senolytics. But with AI, researchers found 21 senolytics candidates in 5 mins by analyzing 4,340 molecules and not to mention saving at least $100k and tons of time.
You all must have known and met a John. Johns are just like us but are only a little more prone to carpal tunnel syndrome. Always remember to smile when you see a John. 👦 Google might be ditching plans for an AI-powered chatbot for GenZ (that shit would have been totally sick but Google is so over!)
🚰 GPT-4's parameter and training details (leaked) - Nothing, just two tech guys writing on Substack (one of which is Patel)
TL;DR => More parameters, A 16 Mixture of Experts (MoE) model over one big model, and a larger training dataset🐶 Unicorn startup Character.AI struggles with waifus and copyrights - Well, you can’t just make a character out of Xi Jinping and expect not to be sued. (Remember Winnie the Pooh?)
🦾 Actors say Hollywood studios want their AI replicas for free, forever. So, they plan to go on strike for now, to ensure machines do not replace them. (being an artist right now must be so scary, constantly being wary of a new tech replacing you)
🍵 Match Cutting: Finding Cuts with Smooth Visual Transitions Using Machine Learning - Here’s an interesting blog by the Netflix Team on how they use ML algos to find appealing Match Cuts for their trailers and promos. (Thanks filterpapi and
)
“A match cut is a transition between two shots that uses similar visual framing, composition, or action to fluidly bring the viewer from one scene to the next. It is a powerful visual storytelling tool used to create a connection between two scenes.”
Source: Netflix Technology Blog 🧠 PoisonGPT: How we hid a lobotomized LLM on Hugging Face to spread fake news (Lobotomy is such a cool word. I last heard it in Shutter Island)
TL;DR => “maLLMware” (yes we made that up) is real and can be easily spread through OSS models.
Translation: Be Cautious, Stay Alert
TILs 🤯
🔖 Worried about your unread emails counting going to the moon? (us too).
Recently, found out that we have Gmail Offline, which allows us to download a replica of the last n days of emails with attachments. 🫡 (this week’s Nibble has a lot of content from a flight, where we utilized this feature)
🌴 Not everything dies from overwatering (considering Nibbler A killed a cute little cactus from extra doses of
lovewater). Mangroves not only flourish in a lot of water but helps balance the ecosystem too.
Mangroves are many species of tree adapted to living in salty water | Source: Sketchplanations
What we have been trying 🔖
2Short.ai: AI-powered short clip generator
Netlify Drop ChatGPT plugin - This plugin allows you to generate code from ChatGPT that can drag and drop to Netlify and deploy. Dope!
Thiggle- API that can categorize, classify or label ‘anything’ (the docs only talk about text. copywriters 🤷♂️) using LLMs
Scrolling Screenshots - Ever got stuck while capturing a full-page screenshot of a website? No more installing Chrome extensions that can be dangerous, just use this tool and download a full-page screenshot in mp4, png, or GIF.
Builders’ Nest 💪🏻
MindOS - An AI agent for every need (*whispers* UrbanAI)
Supertools - A list of all the popular AI tools that you would ever need!
threads-re - Threads private APIs reverse-engineered
dalai - Simple way to run a LLaMA or Alpaca LLM locally (It’s really simple)
Meme of the week 👑
Off-topic reads/watches 🧗
👨🏫 How to pass any first-round interview by
and - Looking for a job, reader? (We know you do, we didn’t add the job section below just like that 😉) Though hiring’s still frozen at many places, you better be prepared before the push comes to shove (just sayin’)👨💼 How to avoid sales fluff in Devtools by
- Working on your next SaaS tool? Follow the Benefit Layers Framework to improve your copy.🏃♂️ Fast by Patrick Collison - Some examples of people quickly accomplishing ambitious things together
👀 The Lucretius Problem: How History Blinds Us by Farnam Street - “… the worst event that has happened in the past surpassed the worst event that came before it. Only the fool believes all he can see is all there is to see.”
✍️ Hey! you, yes you! keep writing if you do. Nobody cares, but keep writing
(P.S. - We didn’t watch a lot since the last edition [pls suggest us some good watches!], so all reading reccs here this week)
Ponder-worthy words 🤔
“Compare yourself to who you were yesterday, not to who someone else is today.”
― Dr. J Peterson
Job Posts 💼
PS: This section is not sponsored, we hand-pick stuff by ourselves, from the week-long scrolling of the web and email reading. We will usually post Global or India-based opportunities.
ML Engineer at Clearcover [Chicago, IL or Remote]
Junior Full Stack JavaScript at Showcase [Remote]
Senior Software Engineer at Bedrock [Remote]
Front End Engineer at LaunchBrightly [Remote]
Blockchain Engineer, Frontend Engineer, and more roles open at NPCI (the team behind UPI) [Mumbai/Hyderabad, India]
From the Author’s Lap(top)
You can reach out to us using our nests on the Internet.
- Aashutosh - Rajasthani Baniya, has a craving for memes, loves *-wares (soft, hard, middle, ALL), can talk in italics, is a part-time cat-whisperer, and can tirelessly work towards the bloom doom of humanity.
- Pushkar, Gujarati Bihari Dikra, loves languages (programming and otherwise) and can be a part-time meme-lord, and a part-time neural network whisperer.
👋🏻 See you at the next one!
Shaka Laka Boom Boom age! 🤣👍