Releases

Explore resources related to releases to help implement AI solutions for your business.

Andrej Karpathy releases concise GPT implementation. Why has he bothered to do this: doesn't he work for OpenAI, at least indirectly? [D] [N]

It's nice to see a concise implementation of GPT, in pytorch, as it is true Hugging Face's Transformer's is excellent, but it is quite difficult to trace. They are trying to build it out constantly with loads of features, so you get lost. I'm wondering Andrej's reason for doing this? His wiki states he works for OpenAI and Tesla is at least affiliated with Openai. Also it's very far from computer vision domain, so why spend the time on an open source implementation and make some guesses on GPT-2/GPT-3. His implementation is easy to follow, which is nice , most reimplementations I see have bugs or are unecessary complex. I am surprised at how basic his trainer code is etc., that he pretty much uses the very standard stuff. Andrej also states 'A vanilla multi-head masked self-attention layer with a projection at the end. I believe I could have just used torch.nn.MultiheadAttention but their documentation is all but absent and code ugly so I don't trust it, rolling my own here.' I usually trust the torch team blindly to come up with a more efficient implementation than I ever could, as don't they stringently test their code, before releasing? so am also surprised by this.

#reddit #andrej #karpathy

[N] NNAISENSE releases EvoTorch (https://evotorch.ai): An open-source Evolutionary Algorithm Library with multi-CPU/multi-GPU support for massive evolutionary experiments!

NaturalGradient•This week

[N] NNAISENSE releases EvoTorch (https://evotorch.ai): An open-source Evolutionary Algorithm Library with multi-CPU/multi-GPU support for massive evolutionary experiments!

Hi everyone! We're excited to release EvoTorch: An open-source Evolutionary Algorithm Library with multi-CPU/multi-GPU support for massive evolutionary experiments! We are researchers and engineers in industrial automation at NNAISENSE and this library is the latest version of the tool we've been using in our own work. It is built on top of PyTorch and Ray, and provides a collection of state-of-the-art evolutionary algorithms, out-of-the-box support for scaling experiments to arbitrary clusters of CPUs and GPUs, advanced tools for NeuroEvolution of any PyTorch module, and direct interfaces to modern logging libraries to track experiments and integrate with existing workflows. Our goal is to make many more engineers be able to use EAs in their workflow, and help researchers do better research. We are happy to have your feedback or questions here or on our Slack channel! EvoTorch is pip installable: pip install evotorch Read the docs: https://docs.evotorch.ai The library is completely open source under an Apache 2.0 license: https://github.com/nnaisense/evotorch

#reddit #nnaisense #releases

Serious B2B businesses will not try to create a solution using AI - This is why. [i will not promote]

consultali•This week

Serious B2B businesses will not try to create a solution using AI - This is why. [i will not promote]

After architecting and developing multiple B2B SaaS platforms and resolving countless challenges, here's why I don't think a proper B2B solution can be developed using AI. You must have senior tech-folks in your teams - even if you choose to leverage AI for expediting some code generation. This isn't theory - this is battle-tested reality. You can use this as a template if you're building one. Core Considerations: Multi-Tenancy Foundation (B2B) Proper tenant isolation at every layer (data, compute, networking) Flexible deployment models (pooled vs. silo) based on customer tier Tenant-aware everything (logging, metrics, tracing) Identity & Security (B2B/Standalone) Enterprise-grade authentication, often with SSO support Role-based access control (RBAC) at tenant level (may need dynamic policy generation for resource access) Audit trails for all system actions (specially if you're in a regulated domain) Client/Tenant Management (B2B) Self-service onboarding with admin approval workflows Automated tenant provisioning/deprovisioning Tenant-specific configurations and customizations Cross-tenant analytics and administration Operational Excellence (B2B/Standalone) Zero-downtime deployments (helps with canary releases) Tenant-isolated debugging capabilities Resource quotas and throttling by tenant tier Automated backup and disaster recovery per tenant Scalability Architecture (B2B) Independent scaling of tenant workloads Resource isolation for "noisy neighbor" prevention Tier-based performance guarantees (SLAs) Dynamic resource allocation Each of these topics can be as complicated as you can think of - depends on the solution you're building. I have seen many seasoned architects and developers struggle also because of their "single-tenant" mindset. Here are some common pitfalls to avoid (B2B/Standalone): Standalone - mindset in database design Hard-coded configurations Lack of context in logging/monitoring Insufficient tenant isolation in shared services (B2B) Missing tenant-aware cost allocation (B2B) You need people great with infrastructure as well. They need to consider: Tenant-aware routing (API Gateway or whatever you're using) Code with isolation when/if required Data storage with proper partitioning Shared services vs. dedicated services strategy There are a number of common problems I have seen people often make. Often it's because of a pressure from high above. But every architectural decision must considered in terms of the solution you're building. In many cases, security cannot be bolted on later, observability must be tenant-aware from day one, operations must scale. This is just the foundation. Your actual business logic sits ON TOP of all this. Now, would you think these can be done by AI? I'll be waiting for that day. :-)

#reddit #serious #businesses

How I Built a $6k/mo Business with Cold Email

Afraid-Astronomer130•This week

How I Built a $6k/mo Business with Cold Email

I scaled my SaaS to a $6k/mo business in under 6 months completely using cold email. However, the biggest takeaway for me is not a business that’s potentially worth 6-figure. It’s having a glance at the power of cold emails in the age of AI. It’s a rapidly evolving yet highly-effective channel, but no one talks about how to do it properly. Below is the what I needed 3 years ago, when I was stuck with 40 free users on my first app. An app I spent 2 years building into the void. Entrepreneurship is lonely. Especially when you are just starting out. Launching a startup feel like shouting into the dark. You pour your heart out. You think you have the next big idea, but no one cares. You write tweets, write blogs, build features, add tests. You talk to some lukewarm leads on Twitter. You do your big launch on Product Hunt. You might even get your first few sales. But after that, crickets... Then, you try every distribution channel out there. SEO Influencers Facebook ads Affiliates Newsletters Social media PPC Tiktok Press releases The reality is, none of them are that effective for early-stage startups. Because, let's face it, when you're just getting started, you have no clue what your customers truly desire. Without understanding their needs, you cannot create a product that resonates with them. It's as simple as that. So what’s the best distribution channel when you are doing a cold start? Cold emails. I know what you're thinking, but give me 10 seconds to change your mind: When I first heard about cold emailing I was like: “Hell no! I’m a developer, ain’t no way I’m talking to strangers.” That all changed on Jan 1st 2024, when I actually started sending cold emails to grow. Over the period of 6 months, I got over 1,700 users to sign up for my SaaS and grew it to a $6k/mo rapidly growing business. All from cold emails. Mastering Cold Emails = Your Superpower I might not recommend cold emails 3 years ago, but in 2024, I'd go all in with it. It used to be an expensive marketing channel bootstrapped startups can’t afford. You need to hire many assistants, build a list, research the leads, find emails, manage the mailboxes, email the leads, reply to emails, do meetings. follow up, get rejected... You had to hire at least 5 people just to get the ball rolling. The problem? Managing people sucks, and it doesn’t scale. That all changed with AI. Today, GPT-4 outperforms most human assistants. You can build an army of intelligent agents to help you complete tasks that’d previously be impossible without human input. Things that’d take a team of 10 assistants a week can now be done in 30 minutes with AI, at far superior quality with less headaches. You can throw 5000 names with website url at this pipeline and you’ll automatically have 5000 personalized emails ready to fire in 30 minutes. How amazing is that? Beyond being extremely accessible to developers who are already proficient in AI, cold email's got 3 superpowers that no other distribution channels can offer. Superpower 1/3 : You start a conversation with every single user. Every. Single. User. Let that sink in. This is incredibly powerful in the early stages, as it helps you establish rapport, bounce ideas off one another, offer 1:1 support, understand their needs, build personal relationships, and ultimately convert users into long-term fans of your product. From talking to 1000 users at the early stage, I had 20 users asking me to get on a call every week. If they are ready to buy, I do a sales call. If they are not sure, I do a user research call. At one point I even had to limit the number of calls I took to avoid burnout. The depth of the understanding of my customers’ needs is unparalleled. Using this insight, I refined the product to precisely cater to their requirements. Superpower 2/3 : You choose exactly who you talk to Unlike other distribution channels where you at best pick what someone's searching for, with cold emails, you have 100% control over who you talk to. Their company Job title Seniority level Number of employees Technology stack Growth rate Funding stage Product offerings Competitive landscape Social activity (Marital status - well, technically you can, but maybe not this one…) You can dial in this targeting to match your ICP exactly. The result is super low CAC and ultra high conversion rate. For example, My competitors are paying $10 per click for the keyword "HARO agency". I pay $0.19 per email sent, and $1.92 per signup At around $500 LTV, you can see how the first means a non-viable business. And the second means a cash-generating engine. Superpower 3/3 : Complete stealth mode Unlike other channels where competitors can easily reverse engineer or even abuse your marketing strategies, cold email operates in complete stealth mode. Every aspect is concealed from end to end: Your target audience Lead generation methods Number of leads targeted Email content Sales funnel This secrecy explains why there isn't much discussion about it online. Everyone is too focused on keeping their strategies close and reaping the rewards. That's precisely why I've chosen to share my insights on leveraging cold email to grow a successful SaaS business. More founders need to harness this channel to its fullest potential. In addition, I've more or less reached every user within my Total Addressable Market (TAM). So, if any competitor is reading this, don't bother trying to replicate it. The majority of potential users for this AI product are already onboard. To recap, the three superpowers of cold emails: You start a conversation with every single user → Accelerate to PMF You choose exactly who you talk to → Super-low CAC Complete stealth mode → Doesn’t attract competition By combining the three superpowers I helped my SaaS reach product-marketing-fit quickly and scale it to $6k per month while staying fully bootstrapped. I don't believe this was a coincidence. It's a replicable strategy for any startup. The blueprint is actually straightforward: Engage with a handful of customers Validate the idea Engage with numerous customers Scale to $5k/mo and beyond More early-stage founders should leverage cold emails for validation, and as their first distribution channel. And what would it do for you? Update: lots of DM asking about more specifics so I wrote about it here. https://coldstartblueprint.com/p/ai-agent-email-list-building

#reddit #built #$6k/mo

I made a Voice AI Automated Testing platform (because I hate making phone calls)

0.5

LemaLogic_com•This week

I made a Voice AI Automated Testing platform (because I hate making phone calls)

As my first New Year’s resolution, I’m excited to officially launch my side project: Testzilla.ai. While designing my Voice AI systems using VAPI, RetellAI, Bland, etc., I quickly got tired of the "Update system, test call flows, repeat" cycle that went with it. The whole point of Voice AI (for me) was that I could get off the phone, not spend even more time on it. So I made some Voice AI agents to test my Voice AI system so I didn't have to keep doing it manually. I showed it to developers friends who got excited and wanted to use it themselves with their systems (and sent me "Take My Money" meme, always a good sign). After hearing this a bunch of times, I decided to make it a platform I could share and easily use on multiple projects, have a simple UI, and let me run tests from my desktop or mobile with a click—and not spend 5-30 minutes of awkward time talking to phonebots in a crowded office. Win. It also has the benefit of being a way for an AI Agency to PROVE to clients that their AI system is working properly, answering questions the right way, NOT answering questions the wrong way, and that any advanced functionality (lookups, appointments, etc.) works properly. Key Features: Multi-Project Management: Simplifies the QA process across a diverse project portfolio, ideal for agencies handling multiple clients. Custom Test Management: Easily create, organize, and track test cases tailored to your project. Run Test Batches: Group and execute test cases efficiently to keep your workflow smooth and organized. Actionable Insights: Get analysis and suggestions that help you fix issues early and improve your releases. Client-Friendly Reporting: Provides clear, detailed reports that make it easy to share progress and results with stakeholders. Developer Tools: Easily manage (receive, email, view, listen, notify) your Transcripts from other systems (VAPI, Retell, etc) without having to create Zapier or Make automations with the provided Webhook URL. More dev tools coming soon, let us know what would make your life easier! I’m launching today and would love to get feedback from this awesome community! If you’re into QA, software development, or just love testing tools, give it a look and let me know what you think. I'll add $20 in credits to your new account so you can try it out risk free, no credit cards required. Here’s the link: Testzilla.ai Looking forward to hearing your thoughts! Cheers, Brian Gallagher

#reddit #made #voice

[N] OpenAI's new language model gpt-3.5-turbo-instruct can defeat chess engine Fairy-Stockfish 14 at level 5

Wiskkey•This week

[N] OpenAI's new language model gpt-3.5-turbo-instruct can defeat chess engine Fairy-Stockfish 14 at level 5

This Twitter thread (Nitter alternative for those who aren't logged into Twitter and want to see the full thread) claims that OpenAI's new language model gpt-3.5-turbo-instruct can "readily" beat Lichess Stockfish level 4 (Lichess Stockfish level and its rating) and has a chess rating of "around 1800 Elo." This tweet shows the style of prompts that are being used to get these results with the new language model. I used website parrotchess\[dot\]com (discovered here) (EDIT: parrotchess doesn't exist anymore, as of March 7, 2024) to play multiple games of chess purportedly pitting this new language model vs. various levels at website Lichess, which supposedly uses Fairy-Stockfish 14 according to the Lichess user interface. My current results for all completed games: The language model is 5-0 vs. Fairy-Stockfish 14 level 5 (game 1, game 2, game 3, game 4, game 5), and 2-5 vs. Fairy-Stockfish 14 level 6 (game 1, game 2, game 3, game 4, game 5, game 6, game 7). Not included in the tally are games that I had to abort because the parrotchess user interface stalled (5 instances), because I accidentally copied a move incorrectly in the parrotchess user interface (numerous instances), or because the parrotchess user interface doesn't allow the promotion of a pawn to anything other than queen (1 instance). Update: There could have been up to 5 additional losses - the number of times the parrotchess user interface stalled - that would have been recorded in this tally if this language model resignation bug hadn't been present. Also, the quality of play of some online chess bots can perhaps vary depending on the speed of the user's hardware. The following is a screenshot from parrotchess showing the end state of the first game vs. Fairy-Stockfish 14 level 5: https://preview.redd.it/4ahi32xgjmpb1.jpg?width=432&format=pjpg&auto=webp&s=7fbb68371ca4257bed15ab2828fab58047f194a4 The game results in this paragraph are from using parrotchess after the forementioned resignation bug was fixed. The language model is 0-1 vs. Fairy-Stockfish level 7 (game 1), and 0-1 vs. Fairy-Stockfish 14 level 8 (game 1). There is one known scenario (Nitter alternative) in which the new language model purportedly generated an illegal move using language model sampling temperature of 0. Previous purported illegal moves that the parrotchess developer examined turned out (Nitter alternative) to be due to parrotchess bugs. There are several other ways to play chess against the new language model if you have access to the OpenAI API. The first way is to use the OpenAI Playground as shown in this video. The second way is chess web app gptchess\[dot\]vercel\[dot\]app (discovered in this Twitter thread / Nitter thread). Third, another person modified that chess web app to additionally allow various levels of the Stockfish chess engine to autoplay, resulting in chess web app chessgpt-stockfish\[dot\]vercel\[dot\]app (discovered in this tweet). Results from other people: a) Results from hundreds of games in blog post Debunking the Chessboard: Confronting GPTs Against Chess Engines to Estimate Elo Ratings and Assess Legal Move Abilities. b) Results from 150 games: GPT-3.5-instruct beats GPT-4 at chess and is a \~1800 ELO chess player. Results of 150 games of GPT-3.5 vs stockfish and 30 of GPT-3.5 vs GPT-4. Post #2. The developer later noted that due to bugs the legal move rate was actually above 99.9%. It should also be noted that these results didn't use a language model sampling temperature of 0, which I believe could have induced illegal moves. c) Chess bot gpt35-turbo-instruct at website Lichess. d) Chess bot konaz at website Lichess. From blog post Playing chess with large language models: Computers have been better than humans at chess for at least the last 25 years. And for the past five years, deep learning models have been better than the best humans. But until this week, in order to be good at chess, a machine learning model had to be explicitly designed to play games: it had to be told explicitly that there was an 8x8 board, that there were different pieces, how each of them moved, and what the goal of the game was. Then it had to be trained with reinforcement learning agaist itself. And then it would win. This all changed on Monday, when OpenAI released GPT-3.5-turbo-instruct, an instruction-tuned language model that was designed to just write English text, but that people on the internet quickly discovered can play chess at, roughly, the level of skilled human players. Post Chess as a case study in hidden capabilities in ChatGPT from last month covers a different prompting style used for the older chat-based GPT 3.5 Turbo language model. If I recall correctly from my tests with ChatGPT-3.5, using that prompt style with the older language model can defeat Stockfish level 2 at Lichess, but I haven't been successful in using it to beat Stockfish level 3. In my tests, both the quality of play and frequency of illegal attempted moves seems to be better with the new prompt style with the new language model compared to the older prompt style with the older language model. Related article: Large Language Model: world models or surface statistics? P.S. Since some people claim that language model gpt-3.5-turbo-instruct is always playing moves memorized from the training dataset, I searched for data on the uniqueness of chess positions. From this video, we see that for a certain game dataset there were 763,331,945 chess positions encountered in an unknown number of games without removing duplicate chess positions, 597,725,848 different chess positions reached, and 582,337,984 different chess positions that were reached only once. Therefore, for that game dataset the probability that a chess position in a game was reached only once is 582337984 / 763331945 = 76.3%. For the larger dataset cited in that video, there are approximately (506,000,000 - 200,000) games in the dataset (per this paper), and 21,553,382,902 different game positions encountered. Each game in the larger dataset added a mean of approximately 21,553,382,902 / (506,000,000 - 200,000) = 42.6 different chess positions to the dataset. For this different dataset of \~12 million games, \~390 million different chess positions were encountered. Each game in this different dataset added a mean of approximately (390 million / 12 million) = 32.5 different chess positions to the dataset. From the aforementioned numbers, we can conclude that a strategy of playing only moves memorized from a game dataset would fare poorly because there are not rarely new chess games that have chess positions that are not present in the game dataset.

#reddit #openai's #language

[N] TheSequence Scope: When it comes to machine learning, size matters: Microsoft's DeepSpeed framework, which can train a model with up to a trillion parameters

Kseniase•This week

[N] TheSequence Scope: When it comes to machine learning, size matters: Microsoft's DeepSpeed framework, which can train a model with up to a trillion parameters

Hi there! Offering to your attention the latest edition of a weekly ML-newsletter that focusing on three things: impactful ML research papers, cool ML tech solutions, and ML use cases supported by investors. Please, see it below. Reddit is a new thing for me, and I've been struggling a bit with it, so please don't judge me too harsh for this promotion. This weekly digest is free and I hope you'd find the format convenient for you. Your feedback is very appreciated, and please feel free to sign up if you like it. 📝 Editorial The recent emergence of pre-trained language models and transformer architectures pushed the creation of larger and larger machine learning models. Google’s BERT presented attention mechanism and transformer architecture possibilities as the “next big thing” in ML, and the numbers seem surreal. OpenAI’s GPT-2 set a record by processing 1.5 billion parameters, followed by Microsoft’s Turing-NLG, which processed 17 billion parameters just to see the new GPT-3 processing an astonishing 175 billion parameters. To not feel complacent, just this week Microsoft announced a new release of its DeepSpeed framework (which powers Turing-NLG), which can train a model with up to a trillion parameters. That sounds insane but it really isn’t. What we are seeing is a consequence of several factors. First, computation power and parallelization techniques have evolved to a point where it is relatively easy to train machine learning models in large clusters of machines. Second and most importantly, in the current state of machine learning, larger models have regularly outperformed smaller and more specialized models. Knowledge reusability methods like transfer learning are still in very nascent stages. As a result, it’s really hard to build small models that can operate in uncertain environments. Furthermore, as models like GPT-3 and Turing-NLG have shown, there is some unexplainable magic that happens after models go past a certain size. Many of the immediate machine learning problems might be solved by scaling the current generation of neural network architectures. Plain and simple, when it comes to machine learning, size matters. We would love to hear your opinions about the debate between broader-larger vs. smaller and more specialized models. Leave a comment Now, to the most important developments in the AI industry this week 🔎 ML Research GPT-3 Falls Short in Machine Comprehension Proposed by researchers from a few major American universities, a 57-task test to measure models’ ability to reason poses challenges even for sophisticated models like GPT-3 ->read more in the original paper Better Text Summarization OpenAI published a paper showing a reinforcement learning with human feedback technique that can surpass supervised models ->read more on OpenAI blog Reinforcement Learning with Offline Datasets Researchers from the Berkeley AI Research (BAIR) Lab published a paper unveiling a method that uses offline datasets to improve reinforcement learning models->read more on BAIR blog 🤖 Cool AI Tech Releases New Version of DeepSpeed Microsoft open-sourced a new version of DeepSpeed, an open-source library for parallelizable training that can scale up to models with 1 trillion parameters->read more on Microsoft Research blog 💸 Money in AI AI-powered customer experience management platform Sprinklr has raised $200 million (kudos to our subscribers from Sprinklr 👏). Sprinklr's “AI listening processing” solution allows companies to get structured and meaningful sentiments and insights from unstructured customer data that comes from public conversations on different websites and social platforms. Xometry, an on-demand industrial parts marketplace, raises $75 million in Series E funding. The company provides a digital way of creating the right combination of buyers and manufacturers. Another example of AI implementation into matching two sides for a deal. Real estate tech company Orchard raises $69 million in its recent funding round. Orchard aims to digitize the whole real estate market, by developing a solution that combines machine learning and rapid human assistance to smooth the search, match the right deal, and simplify buying and selling relationships. Cybersecurity startup Pcysys raised $25 million in its funding round. Pcysys’ platform, which doesn’t require installation or network reconfiguration, uses algorithms to scan and “ethically” attack enterprise networks. Robotics farming company Iron Ox raised $20 million in a funding round. The system of farming robots is still semi-autonomous, the company’s goal is to become fully autonomous. Insurtech company Descartes Underwriting raised $18.5 million. The company applies AI and machine learning technologies to climate risk predicting and insurance underwriting. Legaltech startup ThoughtRiver raised $10 million in its Series A round. Its AI solution applied to contract pre-screening aims to boost operational efficiency. Medtech startup Skin Analytics raised $5.1 million in Series A funding. Skin Analytics has developed a clinically validated AI system that can identify not only the important skin cancers but also precancerous lesions that can be treated, as well as a range of lesions that are benign. Amazon, along with several government organizations and three other industry partners, helped fund the National Science Foundation, a high-priority AI research initiative. The amount of funding is not disclosed. The content of TheSequence is written by Jesus Rodriguez, one of the most-read contributors to KDNuggets and TDS. You can check his Medium here.

#reddit #thesequence #scope:

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

regalalgorithm•This week

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

Hi there, just sharing the latest edition of our AI news digest newsletter! We're just a couple of AI grad students doing this for fun, so hope the self promotion is not too annoying (also, welcome feedback). See it below, and feel free to subscribe. Mini Briefs Robotics, AI, and Cloud Computing Combine to Supercharge Chemical and Drug Synthesis IBM recently demoed a complex system for chemical testing and drug synthesis. The system has an AI component that predicts the results of chemical reactions, and a fully automated robotic experiment setup that runs chemical tests 24/7. Users can access the remote robotics lab online, and IBM can also install the system on-premise. With these tools working together, IBM is hoping to reduce typical drug discovery and verification time by half. AI researchers use heartbeat detection to identify deepfake videos Researchers from multiple groups are tackling the challenge of detecting deepfake videos by analyzing the apparent heartbeat of the people depicted in the video. This is possible, because a person’s blood flow changes their skin color ever so slightly, and this change is often detectable via a process called photoplethysmography (PPG). Because deepfakes are not currently optimizing to generate realisitic heartbeats, temporal or spatial anomalies in PPG signals allow resesarchers to detect deepfakes with a 97% accuracy. Advances & Business This AI Expert From Senegal Is Helping Showcase Africans In STEM \- Adji Bousso Dieng will be Princeton’s School of Engineering’s first Black female faculty. Google’s AI-powered flood alerts now cover all of India and parts of Bangladesh \- India, the world’s second most populated nation, sees more than 20% of the global flood-related fatalities each year as overrun riverbanks sweep tens of thousands of homes with them. Two years ago, Google volunteered to help. Finding magnetic eruptions in space with an AI assistant \- MMS look for explosive reconnection events as it flies through the magnetopause - the boundary region where Earth’s magnetic butts up against the solar wind that flows throughout the solar system. This know-it-all AI learns by reading the entire web nonstop \- Diffbot is building the biggest-ever knowledge graph by applying image recognition and natural-language processing to billions of web pages. Bosch and Ford will test autonomous parking in Detroit \- Ford, Bosch, and Dan Gilbert’s real estate firm Bedrock today detailed an autonomous parking pilot scheduled to launch in September at The Assembly, a mixed-used building in Detroit’s Corktown neighborhood. Create your own moody quarantine music with Google’s AI \- Lo-Fi Player, the latest project out of Google Magenta, lets you mix tunes with the help of machine learning by interacting with a virtual room. Apple launches AI/ML residency program to attract niche experts \- As Apple’s artificial language and machine learning initiatives continue to expand, its interest in attracting talent has grown - a theme that’s barely under the surface of the company’s occasionally updated Machine Learning Research blog. Dusty Robotics CEO Tessa Lau Discusses Robotics Start-Ups and Autonomous Robots for Construction \- Tessa Lau is Founder/CEO at Dusty Robotics, whose mission is to increase construction industry productivity by introducing robotic automation on the jobsite. Concerns & Hype Google Offers to Help Others With the Tricky Ethics of AI \- Companies pay cloud computing providers like Amazon, Microsoft, and Google big money to avoid operating their own digital infrastructure. The Peace Dividends Of The Autonomous Vehicle Wars \- The rapid growth of the mobile market in the late 2000s and early 2010s led to a burst of technological progress. Ethics must be part of the development process’ \- The increasing use of AI (artificial intelligence) in the development of new medical technologies demands greater attention to ethical aspects. Analysis & Policy China’s new AI trade rules could hamper a TikTok sale \- TikTok’s attempt to sell itself and avert a possible US ban may run into some complications. The Wall Street Journal reports that China has unveiled new restrictions on AI technology exports that could affect TikTok. Podcast Check out our weekly podcast covering these stories! Website | RSS | iTunes | Spotify | YouTube

#reddit #last #week

regalalgorithm•This week

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

#reddit #last #week

regalalgorithm•This week

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

#reddit #last #week

Releases

Andrej Karpathy releases concise GPT implementation. Why has he bothered to do this: doesn't he work for OpenAI, at least indirectly? [D] [N]

[N] NNAISENSE releases EvoTorch (https://evotorch.ai): An open-source Evolutionary Algorithm Library with multi-CPU/multi-GPU support for massive evolutionary experiments!

Serious B2B businesses will not try to create a solution using AI - This is why. [i will not promote]

How I Built a $6k/mo Business with Cold Email

I made a Voice AI Automated Testing platform (because I hate making phone calls)

[N] OpenAI's new language model gpt-3.5-turbo-instruct can defeat chess engine Fairy-Stockfish 14 at level 5

[N] TheSequence Scope: When it comes to machine learning, size matters: Microsoft's DeepSpeed framework, which can train a model with up to a trillion parameters

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

[N] Last Week in AI News Digest - Automated chemical synthesis, using heartbeats to detect deepfakes, and more!

how I built a $6k/mo business with cold email

better-genshin-impact

AITreasureBox

vector-vein

RD-Agent

generative-ai-use-cases-jp

instill-core

rpaframework

airspace-hugo

DownEdit

PhoenixGo

obsei

bootcamp_machine-learning

panda-etl

aima-java

Overmind

coca

bytom

aion

studio

practical-python-ai-projects