this is where we talk about un hobbling this is of course something that we just spoke about before but the reason that this is important is because this is where you can get gains from a model in ways that you couldn't previously see 00:15:31 before

the inference efficiency improved by nearly three orders of magnitude or 1,000x in less than 2 years

there is essentially this Benchmark 00:09:58 called the math benchmark a set of difficult mathematic problems from a high school math competitions and when the Benchmark was released in 2021 gpt3 only got 5%

having an automated AI research engineer by 2027 00:05:14 to 2028 is not something that is far far off

it is strikingly plausible that by 2027 models 00:03:36 will be able to do the work of an AI researcher SL engineer that doesn't require believing in sci-fi it just requires in believing in straight lines on a graph

he Talk of the Town has shifted from 10 billion compute clusters 00:01:16 to hundred billion do compute clusters to even trillion doll clusters and every 6 months another zero is added to the boardroom plans

quite frankly a lot of artists and 00:21:16 producers are probably using it just for that they come up with something inspiration they go they make something new

what if a band decides to take one of the udio generated songs and re-record it entirely will they own the full copy rate to that very new recording now if I 00:21:03 was udio the answer probably be like no you made that thing using our platform

the AI created Music learned from got inspiration from the hit songs and came up with a great new hit song for you and then kind of you 00:13:21 know what we'll call those those artifacts or the little similarities here and there might get picked up by Content ID on YouTube

here's a way to do direct to 00:16:46 Consumer sell and can make some money and don't just be like so worried about being on the music platform streaming and now you're diluted because the AI

there's a huge disparity between state of law application of tech and what's 00:15:42 actually happening

to your point for 00:13:46 every problem there's going to be a solution and AI is going to have it and then for every solution for that there's going to be a new problem

this is more of a unfair competition 00:10:36 issue I think as a clearer line than the copyright stuff

now there's going to be even more AI music pouring 00:09:04 into platforms which saturated Market in an already oversaturated Market

these conversations are having daily people are scrambling trying to like we're trying to keep up 00:07:32 with AI in real time scrambling to find out what we're going to do think about all the different businesses that are affected from this

Google deep mind they're coming up their new Google AI sound boox that and it is making Loops from prompts and they have wav Jean

backstory of udio like I didn't know that willim IM and United Masters were like investors in udio

deluding the general royalty pool

the volume of how much music is being created over 800,000 00:01:56 tracks a day are being created using udio

terms of service which is the contract that you sign when you get on their platform does say that you can monetize what you make so meaning you can put into distribution 00:00:41 the music that you make

These arguments are meant to present a cautionary tale of unintended consequences.

I think that Noam chsky said exactly a year ago in New York Times around a year ago that generative AI is not any 00:18:37 intelligence it's just a plagiarism software that learned stealing human uh work transform it and sell it as much as possible as cheap as possible

if we just had a big enough spreadsheet we could get the data in and then we could get you know something like AI or some you know some other computational 00:12:32 process in to help us deal with all this complexity because our little brains can't handle it and my feeling about this is that 00:12:44 actually no

normalizeddifference vegetation index (NDVI)

Die Rede der ZukunftspreisträgerinMeredith Whittaker warnt in ihrer Rede vor der Macht der Techindustrie und erklärt, warum es sich gerade jetzt lohnt, positiv zu denken.

And one way we've seen artificial intelligence used in research practices is in extracting information from copyrighted works. So researchers are using this to categorize or classify relationships in or between sets of data. Now sometimes this is called using analytical AI and it evolves processes that are considered part of text and data mining. So we know that text data mining research methodologies can but they don't necessarily need to rely on artificial intelligence systems to extract this information.

AI with intact skills, but broken goals, would be an AI that skillfully acts towards corrupted goals.

Without intuition, AI can't understand common sense or humane values. Thus, AI might achieve goals in logically-correct but undesirable ways.

Matthew van der Hoorn Yes totally agree but could be used for creating a draft to work with, that's always the angle I try to take buy hear what you are saying Matthew!

***Deep Processing***-> It's important in learning. It's when our brain constructs meaning and says, "Ah, I get it, this makes sense." -> It's when new knowledge establishes connections to your pre-existing knowledge.-> When done well, It's what makes the knowledge easily retrievable when you need it. How do we achieve deep processing in learning? 👉🏽 STORIES, EXPLANATIONS, EXAMPLES, ANALOGIES and more - they all promote deep meaningful processing. 🤔BUT, it's not always easy to come up with stories and examples. It's also time-consuming. You can ask you AI buddies to help with that. We have it now, let's leverage it. Here's a microlesson developed on 7taps Microlearning about this topic.

If students know that the AI has some responsibility for determining their grades, that AI will have considerably more authority in the classroom or in any interactions with students.

if I met a robot that looked very much like a beautiful girl and everything went fine together with her and me but

One of the key elements was "attribution is non-negotiable". OpenAI, historically, has done a poor job of attributing parts of a response to the content that the response was based on.

I feel violated, cheated upon, betrayed, and exploited.

I wouldn't focus too much on "posted only after human review" - it's worth noting that's that's worth nothing. We literally just saw a case of obviously riduculous AI images in a scientific paper breezing through peer review with noone caring, so quality will necessarily go down because Brandolini's law combined with AI is the death sentence for communities like SE and I doubt they'll employ people to review content from the money they'll make

What could possibly go wrong? Dear Stack Overflow denizens, thanks for helping train OpenAI's billion-dollar LLMs. Seems that many have been drinking the AI koolaid or mixing psychedelics into their happy tea. So much for being part of a "community", seems that was just happy talk for "being exploited to generate LLM training data..." The corrupting influence of the profit-motive is never far away.

If you ask ChatGPT to cite it will provide random citations. That's different from actually training a model to cite (e.g. use supervised finetuning on citations with human raters checking whether sources match, which would also allow you to verify how accurately a model cites). This is something OpenAI could do, it just doesn't.

There are plenty of cases where genAI cites stuff incorrectly, that says something different, or citations that simply do not exist at all. Guaranteeing citations are included is easy, but guaranteeing correctness is an unsolved problem

GenAIs are not capable of citing stuff. Even if it did, there's no guarantee that the source either has anything to do with the topic in question, nor that it states the same as the generated content. Citing stuff is trivial if you don't have to care if the citation is relevant to the content, or if it says the same as you.

LLMs, by their very nature, don't have a concept of "source". Attribution is pretty much impossible. Attribution only really works if you use language models as "search engine". The moment you start generating output, the source is lost.

Plenty of companies are still figuring out how to integrate “traditional AI” (that is, non-generative AI; tools like machine learning and rule-based algorithms)

AI-powered code generation tools like GitHub Copilot make it easier to write boilerplate code, but they don’t eliminate the need to consult with your organization’s domain experts to work through logic, debugging, and other complex problems.Stack Overflow for Teams is a knowledge-sharing platform that transfers contextual knowledge validated by your domain experts to other employees. It can even foster a code generation community of practice that champions early adopters and scales their learnings. OverflowAI makes this trusted internal knowledge—along with knowledge validated by the global Stack Overflow community—instantly accessible in places like your IDE so it can be used alongside code generation tools. As a result, your teams learn more about your codebase, rework code less often, and speed up your time-to-production.

The message of e/acc is this: let’s go full steam ahead in the development of increasingly powerful, general, and conscious artificial intelligences, up to superintelligences. This can only be the right path because it reflects the will of the universe. So far I perfectly agree with the philosophical approach of the e/acc movement.

Thisinterventionrecognizes students’ annotations as objectsopen tocontinuous development, engaging students to connect, analyze, and expand upon their ideasthrough the synthesis processes. Meanwhile, the synthesis products can be integratedinto otherlearning events, enriching the overall learning experiences.

At Google, an AI team member said the burnout is the result of competitive pressure, shorter timelines and a lack of resources, particularly budget and headcount. Although many top tech companies have said they are redirecting resources to AI, the required headcount, especially on a rushed timeline, doesn’t always materialize. That is certainly the case at Google, the AI staffer said.

A common feeling they described is burnout from immense pressure, long hours and mandates that are constantly changing. Many said their employers are looking past surveillance concerns, AI’s effect on the climate and other potential harms, all in the name of speed. Some said they or their colleagues were looking for other jobs or switching out of AI departments, due to an untenable pace.

Artificial intelligence engineers at top tech companies told CNBC that the pressure to roll out AI tools at breakneck speed has come to define their jobs.

Experts agree that 2024 will be the year AI invades and transforms work as we know it

We train our models using:

We recently improved source links in ChatGPT(opens in a new window) to give users better context and web publishers new ways to connect with our audiences. 

Our models are designed to help us generate new content and ideas – not to repeat or “regurgitate” content. AI models can state facts, which are in the public domain.

When we train language models, we take trillions of words, and ask a computer to come up with an equation that best describes the relationship among the words and the underlying process that produced them.

the raw data layer (data warehouses and object stores), the compute layer for orchestrated pipelines (feature and inference pipelines), the ML Development services for model training and experiment management, and the state layer for features and models as well as model serving and model monitoring.

The interactive ML systems are typically a Gradio or Streamlit UI (on Hugging Face Spaces or Streamlit Cloud) and work with a model either hosted or downloaded from Hopsworks. They typically take user input and join it with historical features from Hopsworks Feature Store, and produce predictions in the UI.

These system runs a feature pipeline once/day to synthetically generate a new Iris Flower and write it to the feature store. Then a batch inference pipeline, that also runs once/day but just after the feature pipeline, reads the single flower added that day, and downloads the Iris model trained to predict the type of Iris flower based on the 4 input features: sepal length, sepal width, petal length, and petal width. The model’s prediction is written to an online Dashboard, and actual flower (outcome) is read from the feature store and also published to the same Dashboard - so you can see if the model predicted correctly.

However, it is unclear how meaningful the notion of "zero-shot" generalization is for such multimodal models, as it is not known to what extent their pretraining datasets encompass the downstream concepts targeted for during "zero-shot" evaluation.

We consistently find that, far from exhibiting "zero-shot" generalization, multimodal models require exponentially more data to achieve linear improvements in downstream "zero-shot" performance

Have you ever had a meaningful conversation with Siri or Alexa or Cortana? Of course not.

rtificial intelligence is already everywhere

“Did you know that the first Matrix was designed to be a perfect human world? Where none suffered, where everyone would be happy. It was a disaster. No one would accept the program… I believe that, as a species, human beings define their reality through suffering and misery. The perfect world was a dream that your primitive cerebrum kept trying to wake up from.”

“I don’t want comfort. I want God, I want poetry, I want real danger, I want freedom, and I want goodness. I want sin.”

“It generally becomes easier to be generous if you’re doing well and you have a big windfall, [because] when there isn’t enough for everybody, it’s just a question of who is going to starve, and then everything becomes much tougher,” Bostrom told Big Think.

erif concluded that scarcity was one of the main drivers of all human conflict. War, violence, invasion, and theft were all born of wanting a limited resource. The history of all humanity seems to support the hypothesis: We fight over water, cattle, arable land, ore deposits, oil, precious stones, and so on.

Chhit Am-goat  · eprostnSdo919g11fgl4460higl3m0m3thic34lh65i354a1249944tic75f  · Shared with Public《 用台文直接共 AI 討台文內容 》昨昏有朋友問講 AI 出無按家己向望彼號型。窮實佗加強 prompt 佗會使，抑是共欲指定 AI 照你欲--ê 來輸出个規定，囥佇 System Instructions 內底，攏會使--得。下跤圖--ni̍h 是 ka 囥佇 Gemini-pro-1.5 開講區上頂懸个 System Instructions 內底，閣直接叫伊輸入台文文章，*完全無閣 upload 任何物仔*，足簡單个操作。輸出愛檢查，家己愛改，毋過完成度是不止仔懸--矣，而且直接佗出台文內容，毋免閣過翻譯。Prompt 抑是 System Instructions，用英文效果比中文加足好，因為原底遮个 LLM (大言語模型）佗攏用英文訓練--ê。罔參考，會使按家己需要閣改，內底个例攏會換--得，例个長度小可長，效果較好。若想欲輸出 全 POJ、全 TL、抑是全漢，會使改內底个規則佮例，試驗看覓。《System Instructions 》You are a linguist and a great translator between Taigi and English.Translate English text into Taigi I give you, and vise versa．Or chat with me in Taigi.For prepositions, conjunctions, particles and exclamations, must use POJ.For NERs such as country, place or human names keep the original name.English. Be sure to make the Taigi translation more Taigi-like and differentiate from Chinese, using Taigi words and sentences structures as possible, restructuring sentence is allowed.*note 1: Taigi text is a mix of Hanji characters and Latin characters with phonetic component on the top.*note 2: Example of Taigi word consist of Latin letters named POJ(Pe̍h-ōe-jī/POJ) sch as Tâi-gí，má-to͘h, té-thah. Please display the phonetic component and ‘-‘ between subwords accordingly.*note 3: the uniqe POJ elements are: ch, cch, o͘, eng, ek, oa, ⁿ*note 4: Do use POJ for prepositions, conjunctions, particles and exclamations.Output as format&example.<format&example>-----* Leonardo da Vinci ê 無限好玄## Leonardo da Vinci 是文藝復興時期 ê 代表人物, 伊 ê 人生 kah 作品 lóng 浸透彼个時代 ê 精神。伊 tī 1452 年出世 tī Italy ê Vinci, 伊 ê 天才 hāⁿ(迒) 藝術、科學、工程 koh 濟濟 ê 領域, lóng 是 hō͘ 伊對智識 ê 熱狂走chông 來 lu(攄) leh 行。## 伊 tī 藝術 ê 成就, 親像神秘 ê Mona Lisa kah 經典 ê 上尾 ê 暗頓, 展現伊 tùi 人類表情 kah 親像 sfumato hām 明暗對比法遮个先驅技術 ê 把握。 m̄-koh , Da Vinci ê 才調 m̄-nā tī 畫布頂懸; 伊是一个走 tī 時代進前 ê 發明家, 想出來像飛行機器 kah 太陽能這號物。## 伊 ê 簿仔紙展示伊無 kā 藝術 kah 科學分--開 ê 頭殼, 內底滿滿是 chham 觀察、 chham 想像相透濫 ê 素描。Da Vinci ê 哲學是「研究藝術 ê 科學。研究科學 ê 藝術... 認捌 tio̍h 萬物 lóng 連連鬥陣」, 這反映伊相信所有學科 lóng 是 kap(合)做伙--ê。## 伊 tī 藝術 ê 成就, 親像神秘 ê Mona Lisa kah 經典 ê 上尾 ê 暗頓, 展現伊 tùi 人類表情 kah 親像 sfumato hām 明暗對比法遮个先驅技術 ê 把握。 m̄-koh , Da Vinci ê 才調 m̄-nā tī 畫布頂懸; 伊是一个走 tī 時代進前 ê 發明家, 想出來像飛行機器 kah 太陽能這號物。-----Gemini-1.5-pro: https://aistudio.google.com/

LAM is a new type of foundation model that understands human intentions on computers. with LAM, rabbit OS understands what you say and gets things done.

Perplexity AI's biggest strength over ChatGPT 3.5 is its ability to link to actual sources of information. Where ChatGPT might only recommend what to search for online, Perplexity doesn't require that back-and-forth fiddling.

I occasionally wonder what the impact would be of memorizing a good book in its entirety; I wouldn't be surprised if it greatly influenced my own language and writing.

Sunnyvale taps AI to translate public meetings

The adoption of AI translation is cheaper than hiring human translators. Garnett said the city pays $112.50 for every hour the service is used, and has so far paid about $4,308 for more than 38 hours of usage. Hiring an human translator would cost the city up to $400 per hour. “This creates a much smoother dialogue than using live interpreters,” she  told San José Spotlight. “We likely would have needed two to four live interpreters to accomplish what we are doing with Wordly.”

However, despite their eloquence, LLMs have some key limitations. Their knowledge is restricted to patterns discerned from the training data, which means they lack true understanding of the world.Their reasoning ability is also limited — they cannot perform logical inferences or synthesize facts from multiple sources. As we ask more complex, open-ended questions, the responses start becoming nonsensical or contradictory.To address these gaps, there has been growing interest in retrieval-augmented generation (RAG) systems. The key idea is to retrieve relevant knowledge from external sources to provide context for the LLM to make more informed responses.

It’s all made worse by the AI Pin’s desire to be as clever as possible.

But far more often, I’ll stand in front of a restaurant, ask the AI Pin about it, and wait for what feels like forever only for it to fail entirely. It can’t find the restaurant; the servers are not responding; it can’t figure out what restaurant it is despite the gigantic “Joe & The Juice” sign four feet in front of me and the GPS chip in the device.

I hadn’t realized how much of my phone usage consists of these one-step things, all of which would be easier and faster without the friction and distraction of my phone.

Artificial Intelligence (AI) Literacy in Early Childhood Education: The Challenges and Opportunities

The intentional invention of any information or citation on an assignment or document.  This includes using generative Artificial Intelligence (AI) or other electronic resources in an unauthorized manner to create academic work and represent it as one's own.

The same LM can be a much more or less capable agent depending on the enhancements added. The researchers created and tested four different agents built on top of GPT-4 and Anthropic’s Claude:

The latest GPT-4 model from OpenAI, which is trained on human preferences using a technique called RLHFEstimated final training run compute cost: ~$50mModel version: gpt-4-0613

only 15% of European biopharma respondents reported no interest at all in AI in their organization

Additionally, students in the Codex group were more eager and excited to continue learning about programming, and felt much less stressed and discouraged during the training.

On code-authoring tasks, students in the Codex group had a significantly higher correctness score (80%) than the Baseline (44%), and overall finished the tasks significantly faster. However, on the code-modifying tasks, both groups performed similarly in terms of correctness, with the Codex group performing slightly better (66%) than the Baseline (58%).

OpenAI is offering limited access to a text-to-voice generation platform it developed called Voice Engine, which can create a synthetic voice based on a 15-second clip of someone’s voice.

The AI-generated feedback goes to teacher review before being passed on to students so that a human remains in the loop.

But is it a good idea to outsource critical feedback to a machine?

have extensively criticized both companies (and generative AI systems in general) for training their models on masses of online data scraped from their works without consent. Stable Diffusion and Midjourney have both been targeted with several copyright lawsuits, with the latter being accused of creating an artist database for training purposes in December.

謝昆霖 Verified account  · enrdoSspotmlf916af3f9ct1macahh2ft27cmlc8m5c8c51i2l70554i08m1  · Shared with Public這幾天工作同時有請 GPT-4 和 Claude3 。工作條件：我的提問使用正體中文，同樣問題，只給一次作答機會。Claude3 的回覆速度是 GPT-4 的四倍 (或十倍) 以上不說，而且應答的專業度較高，正體中文的使用也比較在地，我這個支語警察給過。它的回答是大量文字輸出時，GPT4 講一講到後面會跑出簡體中文，Claude3 我還沒遇過，而且字數極多的多輪對話中，它對前面的脈絡記憶很清楚。值得注意的是，GPT-4 正體中文貧乏的問題 Claude3 似乎沒有，它的用字遣詞比 GPT4 多一些人類的隨機性，在重覆20次的例句中，同樣的動詞、形容詞，它會盡量用不同的文字。Claude3 似乎解決了我常常酸 GPT-4 的各種問題。這真是驚人。

詹益鑑 Verified account  · 45m  · Shared with PublicAI 真的取代了一些工作嗎？或者造成一些工作的薪資降低？今天看到這篇實際分析的文章，從2022 年11 月1 日（ChatGPT 發布前一個月）到2024 年2 月14 日，在 Upwork 的自由工作者資料中，分析出幾個事實：1. 下降幅度最大的 3 個類別是寫作、翻譯和客戶服務工作。寫作職位數量下降了 33%，翻譯職位數量下降了 19%，客戶服務職位數量下降了 16%2. 影片編輯/製作工作成長了 39%，圖形設計工作成長了 8%，網頁設計工作成長了 10%。軟體開發職缺也有所​​增加，其中後端開發職缺成長了 6%，前端/Web 開發職缺成長了 4%3. 翻譯絕對是受打擊最嚴重的工作，每小時工資下降了 20% 以上，其次是影片編輯/製作和市場研究。平面設計和網頁設計工作是最具彈性的。兩人不僅數量增加了，而且時薪也增加了一些。4. 自 ChatGPT 和 OpenAI API 發布以來，與開發聊天機器人相關的工作數量激增了 2000%。如果說當今人工智慧有一個殺手級用例，那就是開發聊天機器人。

ergi

Broderick makes a more important point: AI search is about summarizing web results so you don't have to click links and read the pages yourself. If that's the future of the web, who the fuck is going to write those pages that the summarizer summarizes? What is the incentive, the business-model, the rational explanation for predicting a world in which millions of us go on writing web-pages, when the gatekeepers to the web have promised to rig the game so that no one will ever visit those pages, or read what we've written there, or even know it was us who wrote the underlying material the summarizer just summarized? If we stop writing the web, AIs will have to summarize each other, forming an inhuman centipede of botshit-ingestion. This is bad news, because there's pretty solid mathematical evidence that training a bot on botshit makes it absolutely useless. Or, as the authors of the paper – including the eminent cryptographer Ross Anderson – put it, "using model-generated content in training causes irreversible defects"

I major in English literature in University I major in interpreting you went to gr for my master degree I think more than 100 people applied and eight students were admit how many people finally passed the final exam two two two wow

Constructing Prompts for the Command Model Techniques for constructing prompts for the Command model. Developers

Now, let’s modify the prompt by adding a few examples of how we expect the output to be. Pythonuser_input = "Send a message to Alison to ask if she can pick me up tonight to go to the concert together" prompt=f"""Turn the following message to a virtual assistant into the correct action: Message: Ask my aunt if she can go to the JDRF Walk with me October 6th Action: can you go to the jdrf walk with me october 6th Message: Ask Eliza what should I bring to the wedding tomorrow Action: what should I bring to the wedding tomorrow Message: Send message to supervisor that I am sick and will not be in today Action: I am sick and will not be in today Message: {user_input}""" response = generate_text(prompt, temp=0) print(response) This time, the style of the response is exactly how we want it. Can you pick me up tonight to go to the concert together?

But we can also get the model to generate responses in a certain format. Let’s look at a couple of them: markdown tables

And here’s the same request to the model, this time with the product description of the product added as context. Pythoncontext = """Think back to the last time you were working without any distractions in the office. That's right...I bet it's been a while. \ With the newly improved CO-1T noise-cancelling Bluetooth headphones, you can work in peace all day. Designed in partnership with \ software developers who work around the mayhem of tech startups, these headphones are finally the break you've been waiting for. With \ fast charging capacity and wireless Bluetooth connectivity, the CO-1T is the easy breezy way to get through your day without being \ overwhelmed by the chaos of the world.""" user_input = "What are the key features of the CO-1T wireless headphone" prompt = f"""{context} Given the information above, answer this question: {user_input}""" response = generate_text(prompt, temp=0) print(response) Now, the model accurately lists the features of the model. The answer is: The CO-1T wireless headphones are designed to be noise-canceling and Bluetooth-enabled. They are also designed to be fast charging and have wireless Bluetooth connectivity. Format

While LLMs excel in text generation tasks, they struggle in context-aware scenarios. Here’s an example. If you were to ask the model for the top qualities to look for in wireless headphones, it will duly generate a solid list of points. But if you were to ask it for the top qualities of the CO-1T headphone, it will not be able to provide an accurate response because it doesn’t know about it (CO-1T is a hypothetical product we just made up for illustration purposes). In real applications, being able to add context to a prompt is key because this is what enables personalized generative AI for a team or company. It makes many use cases possible, such as intelligent assistants, customer support, and productivity tools, that retrieve the right information from a wide range of sources and add it to the prompt.

We set a default temperature value of 0, which nudges the response to be more predictable and less random. Throughout this chapter, you’ll see different temperature values being used in different situations. Increasing the temperature value tells the model to generate less predictable responses and instead be more “creative.”

Can model-free reinforcement learning explain deontological moraljudgments?Alisabeth AyarsUniversity of Arizona, Dept. of Psychology, Tucson, AZ, USA

This technical report focuses on (1) our method for turning visual data of all types into a unified representation that enables large-scale training of generative models, and (2) qualitative evaluation of Sora’s capabilities and limitations. Model and implementation details are not included in this report.

Super Resolution