Рубрика: AI

As we find ourselves amidst the ‘ChatGPT moment’, LLMs stand at the fulcrum of a transformative wave, prompting industry leaders to regard this development as a powerful tool to ‘reduce costs and increase profits’. But, the market doesn’t seem to be unfolding as per the hype. A comprehensive understanding of the infrastructure necessary to maximize the potential of this new domain—including insights into the cost-benefit ratio, pertinent use cases, and the motivations driving organizations to adopt such tools—remains elusive.

On top of that, Gartner’s recent research also forecasts a significant slowdown in enterprise deployments in the general AI space. As highlighted in the study, it is projected that over the next two years, the overwhelming costs will exceed the value that will be generated, culminating in about 50% of the large enterprises abandoning their large-scale AI model developments by 2028.

To get to the crux of reality, AIM Research hosted a roundtable discussion comprising of several AI leaders from different industries working in this space. Here are some key insights that came to light:

• Identifying the appropriate use case with quantifiable business benefits is critical. It involves understanding the technology’s capabilities and aligning them with business objectives.
• Starting with a POC allows businesses to evaluate the potential impacts before scaling up. It is also crucial to be aware of the costs involved in scaling up, including cloud and API usage costs.
• A sensible approach to budgeting would involve allocating more towards improving operational efficiency initially through AI integration to optimize processes, cut costs, and improve service levels, and as the system matures, gradually shift funds towards customer acquisition strategies, utilizing AI to enhance personalization and engagement.
• For AI success, organizations must focus on improving prompt engineering for targeted insights, and excel in data fusion to combine various data sources for more accurate and useful information, promoting collaboration and integration within the organization.
• The future of AI seems to be leaning towards agent technology, where multiple AI agents work together to achieve specific tasks, instead of a single AI entity handling all tasks. These technologies would be industry-specific and would collaborate similarly to a human mind, although achieving this level of integration and function is still a far-off goal.
• Organizations are evaluating both API and open-source options for AI integration, weighing factors like speed to market, customization, and regulatory requirements. While APIs might be favored for pilot projects due to their quick deployment, open-source might be the choice for full-fledged production, offering better audit facilities and customization options.

Thus, a report like this could serve as a vital tool in this process, helping stakeholders to assess the potential costs and benefits associated with different implementation strategies, whether it be through API or open-source pathways. It could clarify the complexities of both direct and indirect costs, facilitating smarter decisions that consider factors such as quick deployment and customization options.

Ultimately, such a report could guide organizations in choosing the most suitable and cost-effective solutions for AI integration.

The post LLM Economics – A Guide to Generative AI Implementation Cost appeared first on Analytics India Magazine.

Annually, Google allocates more than $10 billion to sustain its dominance in the search engine sphere, as revealed by the US Justice Department during an ongoing antitrust trial in Washington. Google’s formidable presence and widespread use pose considerable obstacles for rival search engines to establish substantial footholds. While Google undeniably ranks among the foremost search engines, it may not necessarily be the absolute best. The search engine has its own share of problems, with many opining that the search has become terribly bad over time.

However, the advent of LLM was a silver lining and presented opportunities for others to create differentiation in the search engine space. Back in 2021, Sridhar Ramaswamy and Vivek Raghunathan set out to address the gap in the search engine space and launch an alternative product. In February 2023, Neeva AI launched its search engine powered by generative AI. The LLM-powered search engine challenged Google’s fundamentals and offered an ad-free and privacy-focused search experience.

However, just three months later, Snowflake, the cloud-computing based data company, acquired Neeva AI for an undisclosed amount. Recently, AIM caught up with Sridhar Ramaswamy, who currently serves as the Senior Regional Vice President, ASEAN and India at Snowflake, at the Bengaluru leg of the Snowflake Data Cloud World Tour 2023.

Ramaswamy, who previously worked at Google for over 15 years, and led its USD 115 billion advertising tech division, appeared to be the right man to challenge Google. In an exclusive interaction, he said, competition definitely creates better products. “I was passionate about search and genuinely felt that there could be a better search engine if the focus was just on the user.”

Creating a better search engine

Neeva AI was indeed an ambitious project, given that even Microsoft, with all its might, and GPT-4 integration, has not managed to cause Google any trouble in the search engine space. Ramaswamy’s primary focus was not to break Google’s near-hegemony but to create a better search engine for users. “Yes, it was an ambitious project,” he notes but so was Google.

“But part of the benefits you have as a startup is you can change your techniques when you see an opportunity. And so, while the first two to three years of Neeva were very difficult, both technically and in terms of product adoption, early last year, we saw what was happening with GPT-3 and realised that that could be a technological breakthrough.”

“Users like answers upfront and do not want ten different links to further delve into searching for the correct answer,” Ramaswamy, who also worked at Bell Labs for three years, said. While ChatGPT, initially was touted to be a Google killer, its knowledge was limited to September 2021, back then. “ChatGPT can’t give you real-time data but Neeva can,” Ramaswamy said at that time.

Interestingly, Neeva AI was not the only search engine building on the strength of generative AI and making strides in the search engine space. You.com and Perplexity are among the companies that ventured into the search engine market, each offering unique value propositions. However, the surprising twist came when Snowflake acquired Neeva merely three months later, leaving millions of users who were hoping for a superior product taken aback.

The grim reality

The search engine space could possibly be the toughest space to enter for a startup. Ramaswamy states that becoming the default search engine in Safari, as it turns out, is an incredibly convoluted endeavour. In reality, there is no formal process; it all hinges on Cupertino’s (Apple’s) subjective judgement of your qualifications. “There is no process and it’s pretty tough to create a sustainable business,” the veteran said.

Since its inception, Neeva AI raised USD 77.5 million in funding prior to being acquired by Snowflake. “The weird thing is that we absolutely could have raised another round. Because we created an amazing product in a very hot space.” But, Ramaswamy believes a part of the founder’s job is to see beyond the corner and see what is coming.

Ramaswamy believes Neeva was not on the path to a sustainable valuation. Today, Software-as-a-service (SaaS) companies are getting valued at 10-15 times their revenue. “So to justify our previous USD 300 million valuation, we needed to make USD 20 million in revenue. Even though we might have hit 10 million this year, not USD 20 million and hence we just felt like we were out of time.”

Moreover, the search engine market, unequivocally dominated by Google, paints a disconcerting picture of digital monopolisation. Google’s overwhelming presence stifles competition, curbs innovation, and raises privacy concerns. In fact, Google’s monopoly has not gone unnoticed in the eyes of law. Google will soon be subject to the biggest antitrust lawsuit in over a decade.

Bringing LLM search to enterprise

When the duo realised it was no longer beneficial for the company to carry on in the consumer search space, they decided to make better use of their resources and looked towards enterprise. “We had amassed considerable expertise in the realm of search technology, particularly in cost-effective AI utilisation. We recognised the potential of these assets to effect substantial change and decided to integrate them into an enterprise where their impact could be truly transformative.”

Snowflake became the top choice because the co-founders of Neeva found the company very much sincere in its desire to focus on AI/ML and to deliver great products to its customers. “I had multiple chats with Snowflake including the co-founder Benoît Dageville. We thought it would be like a very natural cultural affinity between the teams and the chance for us to make a big difference. And in fact, we have merged all of the machine learning and AI teams within Snowflake under my co-founder Vivek.”

Integrating Neeva’s search capabilities brings a host of benefits for Snowflake’s customers. “Our focus lies in developing conversational marketplace and catalogue search solutions. Additionally, we also have incredible expertise with language models. These areas are integral to enhancing Snowflake’s core product. As we introduce these functionalities, we anticipate increased customer adoption, thereby driving higher consumption,” he concluded.

The post Why Neeva Pivoted to Enterprise AI with Snowflake appeared first on Analytics India Magazine.

For almost three decades, Google has led the market for making information accessible to the public on the internet on anything and everything — in real time. The curiosity market has experienced a paradigm shift since the rise of AI tools generating content for and on behalf of humans. While Google has long preached ‘helpful content written by people, for people, in search results’, its recent actions say otherwise.

Behind the users’ back, the company quietly rewrote its own rules to acknowledge the rise of AI generated content on the internet. In the latest iteration of the company’s “Helpful Content Update,” the phrase “written by people” has been replaced by a statement that search giant is constantly monitoring “content created for people” to rank sites on its search engine.

The linguistic pivot shows that the company does recognise the significant impact AI tools have in content creation. Despite prior declarations of intentions to distinguish between AI and human-authored content, with this move it appears the company is self-contradictory on its stance on the omnipresent AI generated material on the internet.

Yesterday, 404 Media’s Emanuel Maiberg pointed out that the first picture that pops up if you search “tank man” on Google is not (anymore) the iconic picture of the unidentified Chinese man who stood in protest in front tanks leaving Tiananmen Square, but a fake, AI-generated selfie of the history.

Not on the Same Page

AI powered tools hallucinating is hardly a novel concept. The nature of these models making up stuff is inevitable. While these chatbots rehash content on the internet, the possibility of them churning out false information is imaginable. Hence, in the near future, AI bots double-checking “facts” on the basis of their own previously generated content doesn’t look like a good idea.

Google is one of the leading contributors to fight this phenomenon. At the I/O conference, the company’s executives announced plans to take significant steps to identify and contextualise AI content available on its Search. While measures like watermarking and implementing metadata aims to ensure transparency and enable users to differentiate between AI-generated and authentic images, it can only be applied to images as there is no obvious way to watermark AI-generated text.

Owning its messiah complex, yesterday, Google introduced a bunch of notable features to its AI chatbot Bard, including a way to cross-reference its answers through the “Google it” button. The button, which previously let users explore topics related to Bard’s answer on Google, now evaluates whether Bard’s answers align with or contradict information found through Google Search.

Even more concerning is Bard’s newfound responsibility to fact-check its own AI-generated outputs using Google’s search results — reducing the chances of the response being error free.

Incoming: Internet Collapse

While Google is busy updating its “transparent” policies behind the doors, allowing its search to be flooded with unfiltered AI data, in future AI models including Bard are going to be trained on this data. Hence the risk of unfiltered spam datasets to train these models increases.

As the boundaries of AI replication blurs, the looming question is what happens when AI-generated content proliferates across the internet, becoming the primary source for AI model training? The ominous answer: an impending digital collapse.

Google has issued a statement in the recent past asserting its commitment to fortify search results against spam, emphasising that employing AI-generated content to manipulate search rankings is a violation of spam policies within Alphabet. But the latest updates tell a different story.

Standards continue to evolve as AI proliferates. For now, Google appears steadfast in pushing forward the AI advancements by every means possible — be it by updating policies or adding features to Bard, the company’s sole representative against OpenAI’s ChatGPT.

The company seems to be grappling with the duality of AI-generated content, caught between championing its potential and safeguarding its search results. While Google is navigating the slippery slope, the company’s moves holds the potential to make or mar the future AI-infused digital frontier.

The post Google is Officially Killing The Internet With AI appeared first on Analytics India Magazine.

Since ChatGPT took over newsfeeds last winter, we’ve all seen the rumblings of change caused by generative AI across industries, such as media, art, and education, with mixed results and reactions from veterans in each domain. After many memes and tweets about the poor code generated by the original ChatGPT release, OpenAI came up with the ChatGPT Code Interpreter.

While the plug-in may have temporarily addressed the issue troubled by ChatGPT’s lack in this area, it’s unlikely that anyone is going to switch to ChatGPT from VS Code or Jupyter notebooks any time soon. ChatGPT is still just a chatbot after all, and it doesn’t have all the functionality to rival the power of an IDE or data notebook.

However, this release, among others, raises a few questions.

Can generative AI really have an impact on a technical field like data science? Is it really going to improve the efficiency of data teams? Is generative AI just the topic of the year, or will it leave a lasting impact?

Boost Your Data Workflow with Einblick’s AI-native Notebook

In order for generative AI to truly transform the data science process, it needs to be embedded directly in the notebooks where data professionals are working. That’s the approach taken by Einblick, a new AI-native notebook born out of research at MIT and Brown University.

Einblick is an AI-native data notebook that can write and fix code, create beautiful charts, build and tune ML models, and much more. As modern data teams continue to evolve, they require an ever-increasing level of speed, agility, and flexibility. Earlier this year, Einblick launched their AI agent, Einblick Prompt, which is embedded in every Einblick workspace.

With Prompt, users can build out entire data workflows just using natural language. From data cleaning to exploratory data analysis, model building and tuning, Prompt speeds up every aspect of the data science and data analytics process. With Prompt, Einblick has essentially bottled the power of a Jupyter notebook with the simplicity of ChatGPT.

While many tech companies are seizing the opportunity to provide automated data processing or code generation services through large language models, Prompt comes from a company whose mission has always been simplifying and optimizing workflows for data teams. As such, Prompt offers several unique benefits with that goal in mind. Let’s take a look at them.

Context-awareness: Prompt leverages metadata, such as the formatting of column names, dataset names, and data types, so users don’t need to input paragraphs of information to obtain well-commented and tailored code for their dataset and problem.

Automated data processes: Even if you simply instruct Prompt to “Predict survival” using a specific dataset, Prompt will automatically preprocess your data, check for missing values, split your data into training and testing sets, and display evaluation metrics.

Editable code: Since all of Prompt’s code is generated in a data notebook, you can manually edit and test it immediately. Furthermore, Prompt features a “Change this cell” function that allows you to refine the generated code quickly and intuitively using natural language.

One-click bug fixes: If you encounter an error message in Einblick, you can click the “Fix with Prompt” button, and Prompt will debug your error, display where it modified the code, and provide an explanation of the changes made.

Access to LLMs without API keys: Unlike many other apps that require users to provide their own API keys, Einblick manages all of this for you.

Prompt’s specific value-add seamlessly integrates with Einblick’s other core features. But what are they?

Multimodal workflows: Combine Python, SQL, and interactive components like Charts, Tables, and Filters in the same workspace.

2-D canvas layout: Prototype visualizations and models more efficiently, as workflows can be easily arranged side-by-side, reducing the need to scroll through numerous Python cells.

Fully managed, web-based platform: No more time wasted on configuring your environment or ensuring everyone is on the same page. Presenting and sharing your work has never been easier than with Einblick.

The question of the efficacy and relevance of generative AI in the data space is non-trivial, and it’s exciting to see a startup like Einblick addressing the issue in a meaningful and impactful manner. As a platform built by a data team for data teams, embedding generative AI into the platform simply extends what was already a platform focused on making data teams’ lives easier.

Beware Tech Giants Tacking on Generative AI

Although large companies like OpenAI, Google, GitHub, and Jupyter have been investing in new features to accommodate generative AI, for many, creating an efficient and impactful platform for data science and data analytics workflows is not the core of their business offerings. These companies have certainly created transformative products that we still use to this day, but users should still scrutinize the quality of their new releases.

In March 2023, Google Colab, Google’s web-based Python notebook, announced that AI coding functionality would be added to the notebooks, leveraging a family of Google’s code models, Codey. There has been little news other than previews from Google, but GitHub and Jupyter users can test out the benefits of AI-generated code now.

In fact, GitHub, the platform for version control, took its cloud-based AI tool, GitHub Copilot, out of technical preview just last year in June 2022. Copilot is available in popular code editors like Visual Studio Code and Neovim, providing users with automatic code completion. When using one of these code editors, you can prompt Copilot to offer suggestions by entering code comments, but even then, Copilot tends to generate code one line at a time. If Copilot is not offering quality suggestions, you’ll have to start writing out code manually to get it back on track. Keep in mind, however, that Copilot cannot infer any information about your data, so you’ll spend a lot of time editing any generated code for small things like casing, spelling, hyphens, and underscores. The product certainly seems geared more towards software engineers, given their traditional market, rather than data scientists and data teams. GitHub has announced a chat assistant feature, which is in closed public beta at the time of writing.

Similarly, Project Jupyter, the originator of the Python notebook, created Jupyter AI and Jupyternaut. The former is available anywhere running on an IPython kernel, allowing users to query large language models within Python notebooks to generate code using natural language prompts, as well as providing error syntax explanations. Jupyternaut is available only in JupyterLab and functions as a chatbot within the platform. Users can ask Jupyter AI to generate code using a specific model %%ai chatgpt and then supply a natural language prompt. Similar to Copilot, the lack of context awareness means that much of the generated code is boilerplate, so users will spend time manually editing the code generated. With Jupyter AI, you can also type in a much longer natural language query that provides context to the AI, but that can also be time-consuming. As a relatively new release, features like their error explanations are also a bit buggy but can help users who don’t want to switch between tabs constantly.

Special Use-Case: AI Charts for Everyone

Although applications of generative AI are still being explored, natural language interaction offers many opportunities for data teams and individuals to make parts of their work more efficient. One key example is the space of data visualization. Generative AI can simplify tasks such as adjusting color palettes and label formatting through verbal queries. Leveraging users’ instinct for natural language will save a lot of time compared to the manual configuration demanded by conventional tools. Additionally, rapid prototyping and experimentation are integral to data visualization, and with tools like Einblick Prompt, users can simply ask the AI to replicate and modify specific parts of a chart, accelerating the process and generating multiple versions swiftly.

Using traditional BI tools like Excel and Tableau circumvented the need to program, but fitting the flexibility of code into the complex grammar and syntax of hundreds of preset toggles, settings, and options was a challenge. As a result, users can get bogged down with the numerous steps and limited by the design of the tool. In contrast, AI-driven charting can translate verbalized descriptions into desired charts, offering an entryway to the flexibility of code

In fact, the makers of Einblick, seeing the opportunity in data visualization, recently launched ChartGen AI, a free, standalone app that allows users to go from text to chart in seconds. No account is required. Users just need to upload a dataset or link to a Google Sheet, type in their natural language prompt, and then see their data come to life. From scatter plots to histograms to pie charts and more, ChartGen AI can build anything the user describes.

Find Your Data’s Northstar

There are many exciting advancements in the field of generative AI. However, users must ensure that they are investing in the right tool for themselves and their teams. Not all tools have been designed from the outset for data tasks. Therefore, even though major companies were among the first to offer generative AI tools for coding, they may not necessarily be the best choice. The implementation of generative AI will be influenced by the larger company’s mission, audience, and purpose — all of which may be completely unrelated or, at best, tangential to the work of data teams.

To fully harness the power of generative AI, users should consider alternative tools like Einblick. This is especially true when compared to large tech giants, whose primary focus may not have always been the betterment of data analysts and data scientists.

The post How Generative AI is Revolutionising Data Science Tools appeared first on Analytics India Magazine.

Petnow claims to be able to identify dogs and cats from their snouts Kyle Wiggers 16 hours

Standard pet ID tools, like tags and chips, are imperfect. Tags become easily detached, and not every owner is comfortable with the idea of microchipping their pet. Even those who are comfortable often run into problems with microchips, like chip damage and outdated ID databases.

The challenge inspired Jesse Joonho Lim and Ken Daehyun Pak to launch an app, Petnow, that they claim can identify cats and dogs by scanning their faces. Petnow, which has raised $5.25 million in funding so far from Daedeok Venture Partners and DigiCap at a $24 million valuation, is a participant in the Startup Battlefield 200 at TC Disrupt 2023.

Before founding Petnow in 2018, Lim co-lead a semiconductor startup called Chips&Media, which later pursued an IPO. Pak — who, like Lim, has a doctorate in electrical engineering — had been working as an AI video processing researcher for over a decade before joining Petnow.

At a high level, Petnow works by running a camera-based scan of a pet’s face from a mobile app for Android and iOS. Leveraging AI trained on a set of around 200,000 images of dog and cat snouts, collected both by the Petnow team and sourced from users’ pets, Petnow creates a biometric profile of a pet that’s unique to them.

Lest you be concerned the app accidentally captures, say, a family member standing behind a pet, Petnow claims to use an algorithm to automatically detect and hone in on dogs or cats while cropping out the rest.

For dogs, Petnow records a “nose print.” Yes — a nose print. The startup claims that a dog’s nose is as unique as a human fingerprint and doesn’t change over time, making it a reliable way to distinguish between puppers. For cats, Petnow looks at a cat’s “facial contour,” which Petnow says remains distinctive due to cats’ individual “grooming habits.” (To this reporter, that sounds a little more dubious than a dog’s nose print — but I digress.)

Lim and Pak envision people using Petnow to register their pets without visiting a vet, find missing pets and create “pet IDs” for checking insurance status.

“The pet identification market may not be mature at the moment, but it will eventually become big,” Lim and Pak told TechCrunch in an email interview. “Pet identification technology is a fundamental product that can be continuously used by people, unlike products or services that go viral only for short periods of time. The market has great potential, because pets should have IDs like people, and their data can be backed up to form an ultimate pet platform.”

Image Credits: Petnow

But the question is, does the tech work as advertised?

Petnow claims its algorithms are “99% accurate” at identifying individual cats and dogs. It’s well understood, however, that even the best image-analyzing AI is prone to bias — intentional or no.

For example, at least six people, all Black, have been mistakenly arrested by police using facial recognition technology. Facial recognition algorithms are often trained on datasets that lack a critical mass of Black faces — introducing biases. Or, they’re trained on mugshot databases that contain an overwhelming number of Black faces, many shot in poor and grainy lighting conditions that interfere with the algorithm’s ability to distinguish one face from another.

Setting aside the challenges unique to facial recognition for a moment, in the animal realm, even experts struggle to tell the difference between breeds — let alone animals of the same breed. A recent study involving 5,000 dog experts nationwide found that only a small minority could pick out even one of the breeds identified in the dogs by their DNA.

Petnow claims that its training database is continually growing and that it uses AI to ensure pet photos are taken with the best possible brightness and sharpness. (In the same breath — perhaps anticipating questions about data privacy — Petnow says that it doesn’t provide a user’s or pet’s information to third parties without the user’s consent and offers an option to delete stored data at any time.) And Petnow points to a study co-authored by its data scientists in the journal IEEE Access, which shows that its dog nose-print identifying tech was over 99% accurate at distinguishing between noses.

The study dates back to 2021, though, when the training dataset was presumably smaller. And while Petnow claims it’s working on a companion paper for its cat face-recognizing algorithm, it’s yet to make that research public.

The stakes are high. One can imagine an algorithmic mistake stymieing a family’s search for a missing pet, or causing a vet to pull up the wrong animal’s vaccination records.

Those aren’t imminent threats, to be fair, given Petnow’s relatively slow uptake among pet care providers and shelters. While Petnow has around 70,000 users at present, it’s only signed five undisclosed enterprise and public sector customers in France, Toronto and South Korea (where the company is based).

Petnow is pre-revenue, with a $150,000-per-month burn rate. But it anticipates a contract with Korean domestic and international pet insurers by October and pilots in France and a metropolitan government in Canada for their pet registries.

“Thanks to the pandemic, the pet population has grown more rapidly and people spend more time with their pets … There’s still huge room to grow,” Lim and Pak said. “Government pet registry and affiliation programs with pet insurance providers are currently on the way from South Korea, and we’ll have our product enterprise-ready for the North America and Europe regions soon.”

I only hope that Petnow — and its rivals — deliver on their promises of accurate pet identification. To fall short would be irresponsible; misleading pet owners feels like an exceptionally cruel form of deceptive advertising.

Intel announced advancements in Core Ultra processors, an E-core processor with 288 cores, the 5th Gen Intel Xeon, AI development on Intel Developer Cloud and more at Intel Innovation in San Jose, California on Tuesday, September 19. Intel is working toward its five-nodes-in-four-years plan, with a roadmap of new processors and manufacturing techniques projected out to 2025.

Jump to:

• New Xeon processors include 288-core chip
• Intel Developer Cloud is generally available
• Intel Core Ultra processors and the age of the “AI PC”
• Advancements in manufacturing and chiplets

New Xeon processors include 288-core chip

Intel previewed the 5th Gen Intel Xeon processors (Figure A), which will be available on December 14. The 5th generation will include accelerated performance; for example, the Sierra Forest CPU hit benchmarks of 2.5x better rack density and 2.4x higher performance per watt compared to the 4th Gen Xeon.

Figure A

Granite Rapids, the next processor on the roadmap after Sierra Forest, is optimized for AI workloads, with 2x to 3x performance compared to 4th Gen Xeon, according to Intel’s projections. One Sierra Forest variant holds a remarkable 288 cores and 12 channels of memory. The furthest point on the roadmap is the E-core Xeon code-named Clearwater Forest, which will run on the Intel 18A process node, arriving in 2025.

AI supercomputer built on Intel hardware

Xeon processors will be central to a large AI supercomputer built on Intel Xeon processors and Intel Gaudi2 AI hardware accelerators; Gelsinger calls it the “largest supercomputer in Europe.” Its primary customer will be multimedia generative AI firm Stability AI, the maker of Stable Diffusion.

Intel Developer Cloud is generally available

Starting Sept. 19, Intel Developer Cloud has moved from limited release to general availability. This platform is for AI development, training, model optimization and inference; it also lets developers get hands-on with Intel Gaudi2, 5th Gen Xeon processors and the Data Center GPU Max Series 1100 and 1550. Developers can access an API toolkit within Intel Developer Cloud.

The platform is an easy path to Intel-optimized hardware, software and AI from a developer’s own PC, including up-and-coming hardware and software, said Intel CEO Pat Gelsinger at the Sept. 19 keynote at Intel Innovation. “By the time the hardware is available in volume, you’ve already been working on it for months (or) years,” he said.

Intel Developer Cloud is available in three tiers: free, premium and enterprise. More information on pricing and a full list of features can be found here.

Intel Core Ultra processors and the age of the “AI PC”

Intel’s Core Ultra processors, previously known as the code name Meteor Lake, were at the heart of many of the technologies discussed during the presentation. Core Ultra Processors contain Intel’s first integrated neural processing unit. The NPU is an AI-optimized accelerator and inference performed on the PC, providing for power-efficient AI acceleration and local inference on the PC; put another way, it allows applications to take advantage of AI performance offline.

SEE: Intel wants to solve the AI computing skills gap (TechRepublic)

Core Ultra came about due to several of Intel’s hardware advancements, namely the Foveros packaging technology manufacturing technique and the Intel 4 process node with 3D high-performance hybrid architecture.

Acer CEO Jerry Kao demonstrated Intel Core Ultra in an upcoming Acer laptop using Intel-developed AI libraries and the Intel-led open source AI toolkit OpenVINO.

“We see the AI PC as a sea change moment in tech innovation,” Gelsinger said.

Microsoft is using Core Ultra in its Windows 11 PC and plans to use its AI capabilities in upcoming Copilot features.

Core Ultra is exciting because it brings a CPU, GPU and NPU all together, said Gelsinger. “Our NPU will enable AI developers to take advantage of the standard software and framework for AI development and hugely expand the applications for edge deployment,” he said.

Intel Core Ultra will be available starting Dec. 14, 2023.

Advancements in manufacturing and chiplets

Intel’s five-manufacturing-nodes-in-four-years process development program is progressing rapidly, Gelsinger revealed. Each manufacturing node defines a certain semiconductor manufacturing process. The five nodes in the program are referred to as 7, 4, 3, 20A and 18A.

In particular, Gelsinger showed an Arrow Lake processor based on the 20A node. This node is remarkable because it could push Intel ahead of its competitor, TSMC, in terms of rapidly developing new chipmaking techniques and technology.

The next and last node in the program, 18A, is on track for the second half of 2024.

Update to the Universal Chiplet Interconnect Express specification

Last year, Intel announced the founding of the Universal Chiplet Interconnect Express, an industry specification consortium for next-generation chiplets. The group now has about 120 members and has produced its first test chips, code named Pike Creek. On September 19, Intel demonstrated a novel multi-chiplet package based on UCle interconnects.

Subscribe to the Innovation Insider Newsletter

Catch up on the latest tech innovations that are changing the world, including IoT, 5G, the latest about phones, security, smart cities, AI, robotics, and more.

Delivered Tuesdays and Fridays Sign up today