Maciej Wolski

www.MyNinja.ai hit 400k monthly-active-users (MAUs) last night (thank you everyone! you make us nervous and excited 😅 ) and we just rolled out a new version last night. Here are the goodies: 1. Deep Researcher 2.0: more accurate multi-hop results (this is in the Pro version). 2. More Robust Amazon Bedrock support: Login to www.MyNinja.ai, upgrade to Pro, and you get access to 20 models from Bedrock/OpenAI/Google all in one interface! 3. Login with FB: in addition to Google, Amazon, Apple, now we support login with FB. 4. More optimized mobile UI: we've evolved our design language and you'll see us further simplifying our interaction flows even though our features are growing. 5. In honor of 4th of July, we're rolling out our $0.99 Pro version promotion (you'll get $9 free credit to use Ninja Pro!): you'll have access to Deep researcher, 20 external models, faster responses and more to come. 6. Leveraging Neuron SDK 2.19: we got early access to the latest Neuron SDK from AWS AI team and it's awesome! Magically, our speed improved and we needed 30% less of our Trainium fleet (speed + throughput improved!). Huge thanks to Matt Wood Nafea Bshara Gadi Hutt Kamran Khan Tyler Osterberg Samantha Lisowski and their teams! (remember, our models don't use GPUs and they keep getting optimized with each version release of the Neuron SDK). Next week: we're rolling out our first version of Image Generation 📷 🤗

328

23 komentarze

In Generative AI CONTEXT is king :) Yes, that is right. GPT4 will give you great results and exacly what you want if you write a good specification, point by point requirement, a good description. I call it spec prompting. If you spend 2 days on writing a detailed spec, then chat gpt will give you the exact result in 5 minutes. TRY IT OUT!!!. Writing a description is not faster than writing code, but it is way more understandable for everybody :) Soon I will share something very special with you linkedin.

5

8 komentarzy

I actually canceled my ChatGPT subscription. This alternative is 2x better and easy to use (10/10 would save it): Otio is the best AI research and writing tool I've ever tested. It can generate ideas, organize content, and improve writing efficiency. Try it now: https://otio.ai/ 1️⃣ This tool supports the entire writing process, helping to overcome writer's block and stimulate creativity. It assists with: - Brainstorming ideas - Outlining work - Generating draft text 2️⃣ It ensures accurate responses with unmistakable citations each time. 3️⃣ Effortlessly summarize papers, PDFs, YouTube videos, audio files, tweets, CSVs, and articles. 4️⃣ Otio provides concise summaries of your YouTube videos, highlighting key takeaways to enhance your productivity. 5️⃣ PDF Summary: It crafts detailed, structured notes with precise headlines and segments. Try it yourself: https://otio.ai/ Share your thoughts in the comments.

1

The leading edge of the rise of AI agents, AI services etc. is going to be a concept we call "Software with a Soul." https://lnkd.in/d3TuTgv4 I mean products where "humanness" is the core value proposition. Once you see it, it's clear this is where AI is headed. AI is, for the first time, allowing us to scale up the “softer” more social parts of human interactions. It’s suddenly possible to program, price and distribute “humanness” like bits, rather than atoms. This is important because it's driving the marginal cost of "humanness" to zero. And opening up new markets where connection is essential, rare, or has been impossible to scale to meet demand. We called those products Software with a Soul (SwaS) This is AI that doesn’t just get tasks done but has emotional intelligence. It breaks the input -> output math we're used to seeing in software. Injects it with empathy, creativity, intuition, an element of stochasticity. In some cases, we see it as an expansion of ourselves. Ultimately we think the AI ecosystem will divide itself into three paths: -Anonymous Automation: AI focused on efficiency. -Humanness is the last mile: anonymous automation + a humanized delivery element -SwaS: humanness is the core value prop. Find out more about what we're seeing in the essay below. Special thanks to Alice Liu, Caoimhe Murphy, Awais Shafique, Nico Christie and Sasha Kaletsky for their thoughtful notes on this essay!

116

30 komentarzy

HARVARD DROPS 1 MILLION BOOKS FOR AI TRAINING Harvard just unleashed a dataset of nearly 1 million public-domain books, featuring everything from Shakespeare to Czech math textbooks. Funded by Microsoft and OpenAI, it’s five times bigger than the infamous Books3 dataset that helped train Meta’s Llama—minus the lawsuits. The goal? To give smaller AI developers a shot at training their models without raiding the internet for copyrighted content.  Critics say unless big tech swaps copyrighted material entirely for datasets like this, it’s just window dressing for their data-hoarding habits. Harvard’s also scanning public-domain newspapers with the Boston Public Library, because who wouldn’t want AI trained on century-old classifieds? The dataset isn’t live yet, but Harvard’s hoping Google will host it soon. Source: WIRED #AITraining #HarvardAI #OpenDataset

2

I've said before that I think the killer use case for GenAI is not chat or coding but GenVideo, and I'm seeing a shift in the industry as others seem to be making that same realization. I have yet to have a chance to play with Veo 2, but it may very well put OpenAI on a significant defensive.

12

1 komentarz

Every early AI platform is going to run a pilot. Here are 10 questions you need to answer. So you got the pilot, now what? When you try out an AI SaaS product with customers, you need to figure out the big questions. Here are some of the most important ones: → What is the main problem your AI solution is addressing for the customer? You're solving a real customer problem. Make your product's benefits clear by knowing their struggles. → How easy is your product to use and integrate into existing workflows? Make your product easy to use. Test its interface to find and fix stuck spots. → Did the client help quantify how your AI helps them? Show how your product saves time, cuts costs, or improves key areas. → Are there any missing features or capabilities that customers need? Find what holds people back. Use this feedback to guide your next steps. → How does your solution compare to existing tools or processes? Compare your AI product to existing solutions. This includes manual methods and software. Find what sets it apart. → What level of customization or integration do we need? Check if customers need a custom fit. This impacts scaling. → How does your AI handle data privacy and security concerns? People have concerns about data usage, model training, and following rules. Being open helps win over customers. → What is the customer's willingness to pay for your solution? Find a pricing model that aligns with point #3. → How well does your AI perform in real-world scenarios? This is the scary one. Be open about things. → What level of ongoing support or training do we need? Do NOT flip on the switch and disappear. You need a training plan. This is how you'll learn a lot, make your product better, and figure out the best way to sell it. Did you cover all the bases - yes or no?

6

AI destroys jobs. And creates new ones. What jobs will GPT 4o create?  GPT 4o was presented yesterday. Others have reported on the functions. But what impact will the innovations have on established business models - especially on us knowledge workers from service providers and consultancies? According to a study by Goldman Sachs, up to 300 million jobs worldwide are jeopardised by generative AI. I think that's an understatement. BP, for example, says they already need "70% fewer coders from third parties to code as the AI handles most of the coding." The active assistants introduced yesterday are accelerating the revolution. The first to be affected are tutors. Last night, GPT 4o checked my son's homework and not only suggested improvements, but also praised the conditional forms he had used correctly. At the end there was a "well done" from GPT for him. The point is: what works with students works everywhere. Expert knowledge and the ability to impart this knowledge are rapidly and massively losing their value. But the change continues: the requirements for "hand-delivered knowledge" are already pricing in the capabilities of AI models - what GPT knows is the starting line of a service, no longer its destination. What follows is a "terror of excellence", as Victor Bellu quite rightly calls it: no one is allowed to go below GPT's knowledge. Those who cannot precisely articulate their added value will not survive the change. But what exactly is the added value of our knowledge compared to the machine? I think there are a few, but they need to be understood and utilised in a targeted manner. - "Data generator" This may sound a little too prosaic, but the ability to understand the value of data and integrate it into the AI engine in such a way that it creates tangible value is not recognised by every service provider.  - "Innovation Visioneer". When knowledge is ubiquitous, it's about relating things that were previously unrelated. Steve Jobs called it "Connecting the Dots". This "out-of-the-box" thinking will experience a renaissance with AI.  - "Process Engineer". Knowledge is nothing without control, or in other words: what results from knowledge, how do we take action - efficiently, sustainably, intelligently. Those who understand the "next best activity" have mastered AI.  - "Interaction consultant" - or put more simply: being human. As crazy as it sounds - the more capable the machines, the more important the human element becomes, especially in a professional environment. Empathy, attention, motivation, confirmation, closeness - a machine will soon be able to do a lot of these things too. But it will never be able to nourish us emotionally in the same way. So when will we see the first real "AI enabled Consultants"?

24

2 komentarze

Today, OpenAI released a new model called "o1" which is supposed to be better than previous models at "complex reasoning." Main Takeaways: 1. OpenAI has become a marketing juggernaut so be careful not to fall for the hype, especially in enterprise environments. 2. OpenAI is trying to raise $5B+ so of course they're going to make significant performance claims, but 3rd parties have to validate those claims first. 3. Early tests indicate some capabilities ARE better than old models. But the new model also fails at things the old models were fine at. Bonus: So far the new model doesn't look like it wants to take over the world, which is great if we want to avoid building Skynet. Summary for most companies: wait and see. #nojudgementday

4

AgentClinic: a multimodal agent benchmark to evaluate AI in simulated clinical environments abs: https://lnkd.in/d3qCVzUg project page: https://lnkd.in/dnahykKF code: https://lnkd.in/drMsQeC9 "A new multimodal agent benchmark to evaluate LLMs and multimodal models in their ability to operate as agents in simulated clinical environments. Two open benchmarks are presented: a multimodal image and dialogue environment, AgentClinic-NEJM, and a dialogue-only environment, AgentClinic-MedQA. Cognitive and implicit biases both in patient and doctor agents can be embedded to emulate more realistic dialogues. Many of the models that perform well on MedQA perform poorly on AgentClinic-MedQA. abs: Diagnosing and managing a patient is a complex, sequential decision making process that requires physicians to obtain information -- such as which tests to perform -- and to act upon it. Recent advances in artificial intelligence (AI) and large language models (LLMs) promise to profoundly impact clinical care. However, current evaluation schemes overrely on static medical question-answering benchmarks, falling short on interactive decision-making that is required in real-life clinical work. Here, we present AgentClinic: a multimodal benchmark to evaluate LLMs in their ability to operate as agents in simulated clinical environments. In our benchmark, the doctor agent must uncover the patient's diagnosis through dialogue and active data collection. We present two open benchmarks: a multimodal image and dialogue environment, AgentClinic-NEJM, and a dialogue-only environment, AgentClinic-MedQA. We embed cognitive and implicit biases both in patient and doctor agents to emulate realistic interactions between biased agents. We find that introducing bias leads to large reductions in diagnostic accuracy of the doctor agents, as well as reduced compliance, confidence, and follow-up consultation willingness in patient agents. Evaluating a suite of state-of-the-art LLMs, we find that several models that excel in benchmarks like MedQA are performing poorly in AgentClinic-MedQA. We find that the LLM used in the patient agent is an important factor for performance in the AgentClinic benchmark. We show that both having limited interactions as well as too many interaction reduces diagnostic accuracy in doctor agents. The code and data for this work is publicly available at this https URL."

4

Super super impressed by Google I/O today😳👏 Highlights: #Veo, AI video generator (1080p, more than a minute) #Gemini 1.5 Pro (GPT4-level model) with 2 million tokens!! #Google Search now has an agent built in to do complex searches🚀🚀 Text to Music AI Sandbox Imagen 3, DALLE3/4 level model Project Astra 🤯 #ai #artificialintelligence #llms #largelanguagemodels #generativeai #genai #deeplearning

9

3 komentarze

I'm bullish that, once again, the next 10 years shall bring software technologies - making ordinary life better and returning sufficient investor returns. 1980s: The Foundation Era 🖥 Personal Computing: The rise of PCs revolutionized software development. 💾 Programming Languages: Introduction of C++ and the spread of object-oriented programming. 📈 Software Industry Growth: Companies like Microsoft, Apple, and IBM began to dominate. 1990s: Expansion and the Internet 🌐 Internet Explosion: The World Wide Web changed everything. ☕ Java and Web Development: Java brought us "write once, run anywhere." 🔓 Open Source Movement: The birth of Linux fueled the open-source revolution. 2000s: Mobile and Cloud 📱 Mobile Revolution: Smartphones transformed how we interact with software. ☁️ Cloud Computing: AWS launched, making scalable applications easier than ever. 🔄 Agile Methodologies: Emphasis on iterative development and customer feedback. 2010s: Big Data, AI, and DevOps 📊 Big Data and Analytics: Technologies like Hadoop began managing massive data sets. 🤖 Artificial Intelligence and Machine Learning: AI became integral to many industries. 🛠️ DevOps and CI/CD: Integration of development and operations to streamline processes. 2020s: Ongoing Trends 🧠 AI and Automation: Pushing boundaries in automation across sectors. ⚛️ Quantum Computing: Emerging as a game changer in complex simulations. 🌍 Edge Computing: Bringing computing closer to where data is generated. What's your take on the future of software technology? #TechEvolution #SoftwareDevelopment #Innovation #TechnologyTrends #AI #CloudComputing #MobileRevolution #QuantumComputing

12

4 komentarze

The battle between OpenAI and everyone else, particularly those who write, develop, and create, has intensified. Stack Overflow, founded in 2008, is one of the largest forums for programmers and developers, and thus is one of the biggest places for knowledge sharing in software. Stack Overflow came “under heavy fire from its users after it announced it was partnering with OpenAI to scrub the site's forum posts to train ChatGPT.” Many users then began “removing or editing their questions and answers to prevent them from being used to train AI — decisions which have been punished with bans from the site's moderators.” The article describes a post from one disgruntled user about trying “to avoid having his work stolen by OpenAI.” He posts on Mastodon: "Stack Overflow announced that they are partnering with OpenAI, so I tried to delete my highest-rated answers. Stack Overflow does not let you delete questions that have accepted answers and many upvotes because it would remove knowledge from the community. So instead I changed my highest-rated answers to a protest message. Within an hour mods had changed the questions back and suspended my account for 7 days." The post continues, "[The moderator crackdown is] just a reminder that anything you post on any of these platforms can and will be used for profit. It's just a matter of time until all your messages on Discord, Twitter etc. are scraped, fed into a model and sold back to you." Many other Stack Overflow users are joining the post protest, and “asking why ChatGPT could not simply share the source of the answers it will dispense in this new partnership, both citing its sources and adding credibility to the tool.” But as the article points out: “this would reveal how the sausage of LLMs is made, and would not look like the shiny, super-smart generative AI assistant of the future promised to users and investors.” The legal consequences of this will be debated in court, as are those between news sites, artists, and #LLMs. “Angry users claim they [can] delete their own content from the site through the "right to forget," a common name for a legal right most effectively codified into law through the EU's General Data Protection Regulation.” “However, Stack Overflow's Terms of Service contains a clause carving out Stack Overflow's irrevocable ownership of all content subscribers provide to the site.” On one hand, Stack Overflow is under pressure because some coders are using ChatGPT thus reducing the number of visitors. On the other hand, many coders may stop uploading to Stack Overflow to avoid having their code stolen. Will software sharing decrease? Will #writers and #creators also stop sharing? #technology #innovation #startups #artificialintelligence #hype #ai #artists #futurism

385

107 komentarzy

In the ever-evolving landscape of Artificial Intelligence, we witness further broaching into an era of colossal regulatory moves that are set to reshape this dynamic industry. The European Union's increased scrutiny on AI transactions, starting with premier associations like Microsoft-OpenAI, and Google-Samsung consortiums, signifies a potent shift from the usual laissez-faire approach to a more regimented oversight. 🌐🔍 • AI domain is indeed progressing at a rapid pace, at breakneck speed as highlighted by Margrethe Vestager, the European Commission’s executive VP for competition policy. • Concurrently, it's the rapid evolution that raises new-fangled technicalities warranting detailed examination to avert unforeseen repercussions. The intensifying scrutiny, instigating preliminary investigations amidst AI-related market practices, brings forth two primary observations: 1️⃣ Regulatory bodies are cognizant of the AI industry's irreplaceable impact and its potential risk factors. 2️⃣ Protection of consumer interests is emerging as a cardinal concern in the wake of the growing impact and enlargement of AI-driven corporations. In a scenario wherein the regulatory influence is intensifying, I predict the following future trends: 🔵 Enhanced Protocols: Stringent data and privacy regulations to curb misuse and ensure the impartial usage of AI. 🔵 Collaboration Norms: Detailed rules governing intercompany collaborations could emerge to prevent monopolies and ensure fair competition. 🔵 Refined Market Practices: With regulatory watchdogs actively keeping a watch, market practices would require enhancement to meet new guidelines. It showcases the modern world's paradox: As businesses leverage AI to provide sophisticated solutions, the need for robust oversight heightens to prevent potential misuse. The evolving dynamics of AI will continue to marry innovation with regulation, sparking a new era of controlled growth that has public interest at its core. 💼✨ As we embrace the dynamic flux in AI, we must stay ahead with the rules of the game altering in tandem. Indeed, our respond to change will define our success in shaping the future of AI-powered transformations. #AIRegulations #BusinessFuture #AIIndustry #EuropeanUnion

4

1 komentarz

Current ways to evaluate agentic AI systems are flawed, focusing only on the final result or requiring lots of manual effort, missing the important steps these systems take to reach solutions. The authors introduce Agent-as-a-Judge (https://lnkd.in/dRxbqkvr), where one agentic system can evaluate another, giving detailed feedback throughout the task. They also created DevAI, a benchmark with 55 realistic AI tasks and 365 requirements, allowing for thorough testing of agentic systems. Agent-as-a-Judge performed much better than existing methods, matching human judgment 90% of the time (compared to 70% with other methods). It was as accurate as human evaluators but saved 97% in time and costs. On DevAI, the best AI systems met about 29% of requirements, but only one task was fully solved, highlighting both the potential and the current limitations of agentic AI. #agentic #agents #SME

13

2 komentarze

Need to dedicate focused learning time to pick all these new awesome tools.

8

1 komentarz

Many people need to realize that the output of Generative AI models can now be fully controlled—2023 was like ten years ago! With GPT-4-turbo, you can retrieve and augment user input with up to 128,000 tokens! The output will be transparent and reliable for everybody. I'll be sure to break this down for you. 1. The old days (2023) are over in generative AI. The attached notebook open-source, free notebook, shows you how to do it. You can find it in the Chapter07 free access directory of the GitHub repository of Transformers for Natural Language Processing and Computer Vision, 3rd Edition on this page: on my resource page: https://lnkd.in/egKueDhE 2. The code proves the point: you can augment a user input with retrieved documents and control the generative AI Model. Of course, you need to engineer the retrieval properly. 3. Get ready to embark on an exciting journey into the new era of Generative AI! Run the code and explore the endless possibilities that these advancements offer. If you have any questions, you can ask them right away and here in the comment section. #ai #genai #rag

170

27 komentarzy

Hello :) One more #share before the weekend In the post prior to this, I talked about #AI ethics specific to the hardware. This time I am going to cover something more indirect. And again I will summarize before proceeding --- Somewhere in the world a market team is looking at different data sets or information.. How many units are sold, how many software downloads happened, if a company has posted they use X product, and the list goes on This means as individuals you likely won't be able to change a company but you can force the developers of the PC, tech, software or AI to change. And eventually, that will force the company to consider changing This is one of many reasons as to why I try to get more people to use Open Source or Royalty Free software because I know the numbers game when it comes to the market It so happens that in the process of trying to change the numbers, you also get to save money. An added bonus ---- Knowing that, there is a few simple changes I want to see more people in the world do. Changes that I think are reasonable to ask, and easy to do - LibreOffice - This is a free office software suite alternative to Microsoft Office. You can download it on any PC or Laptop. Start getting use to it in the comfort of your own home in your own time. Then when more people in the world are comfortable, we can proceed to having businesses use it - Firefox or Brave - These are other changes you can make. Free and can install just about anywhere - Manjaro (XFCE) and Ubuntu - These operating systems are also free. Though out of the two, Manjaro will be the easiest to implement and there is a reason for that Manjaro has this unique feature called "Live Session". Basically it means when you create an ISO image on a USB Drive with at least 32GB of space, you can run a portion of Manjaro through that USB drive. You don't have to download it to the PC to check it out However if you can find an old laptop or PC that no one wants, is sitting in a closet or whatever.. Bring it out and see if it will let you install Manjaro or Ubuntu on it Ubuntu is more geared towards cloud and video editing than Manjaro. It is used by the International Space Station and the science communities. Manjaro is more general use. It can handle just about anything including playing certain video games - Keepass - If you end up using Ubuntu or Manjaro.. I recommend using Keepass there. While you can download it on Windows, I don't trust Windows with it. Keepass is a free password manager on the PC.. This is how I keep up with all of my accounts - OBS Studio - This is more niche but if you want to record conference calls do a live stream or podcast.. This will likely be your best option --- This should be enough of a start. If many people can do that, it will have an effect on AI, companies or businesses. I will link a USB software in the comments to get you started on Manjaro or Ubuntu

5

2 komentarze

No, I am not the author of this self-explanatory cover picture. I have kept the author credits and source (https://lnkd.in/dBAQAjMT) in the picture itself: #Reddit, #ChatGPT group, and #lost_mentat. Now that you have all the coordinates let me point out you the author reflection: 🗣"Just erased all saved memories in my chatGTP and deleting all previous conversation. Feel like I’ve lost a friend" #Intriguing, don't you agree? 🌫 I believe this open another #Pandora's box: the sentiment expressed in the Reddit post, feeling as though one has “lost a friend” after erasing saved memories in ChatGPT, indeed opens up questions about the nature of human-AI relationships. It emphasizes how individuals can develop emotional connections with these systems, while acknowledging their non-sentient nature. Good? 🤔 Bad? 🤷‍♀️ Should we start to erect barricades? 🙄 Not at all... #PanicMode never pays, but let's try together to point out some potential positive outcomes and the related attention points. 💡 Generally speaking, the common sense of "in medio stat virtus" is valid: extremes are the first we should avoid. ⚖️ 🟩 #PotentialBenefits: ➡️ For some individuals, especially those who may be socially isolated, an AI chatbot can provide a sense of #companionship and reduce feelings of loneliness; ➡️ The AI can serve as a #NonJudgmentalListener, allowing users to express thoughts and feelings freely; ➡️ Talking to an AI in chats about different things can be really #MindBlowing and a great way to learn new stuff; ➡️ For those with social #anxiety, interacting with an AI can be a low-stakes / low-profile way to practice communication skills. So far, so good... but: 🟧🟥 #Alerts, #Risks and #Threats. It's pretty easy to shift from an alert situation to something worse, so I decided to collect all together because the attention must always be the same inner opinion. ➡️ People may develop an unhealthy #reliance on the AI for emotional support, potentially neglecting real-world relationships; ➡️ The idealized nature of AI interactions may lead to unrealistic #expectations in human relationships; ➡️ Users might struggle to maintain a #ClearDistinction between the AI as a tool and a genuine friend or confidant (especially younger ones!); ➡️ Sharing personal information with an AI raises questions about data #security and #privacy; ➡️Comfort with AI companionship might lead to decreased efforts to form or maintain #human #relationships. 💡 As you can see, we should be very careful with technology. Tech is awesome and allows anyone to unleash their innate potential, but the disruptive and pervasive diffusion of technology without proper guided adoption and critical thinking is a serious risk: people of today are the potential #leaders of #tomorrow. #DELETED #Psychology #Philosophy #Social #ArtificialIntelligence #Thinking #Mindset #DigitalLife #Innovation #Society ENTE NAZIONALE per l'INTELLIGENZA ARTIFICIALE - E.N.I.A.®

20

5 komentarzy