Rumours of an Apple AI product have spread rapidly following release of four small language models to developers.

Open source and free to use, ‘Open-source Efficient Language Models’ or ‘OpenELM’ are a set of efficient text generation models available for Apple developers to begin experimenting. So far, the low-key release online without major fanfare has prompted only a handful of downloads.

Unlike Microsoft, Google and Facebook, many believe Apple are working on ‘on-device AI’ models that would not involve sharing large quantities of data with cloud-based platforms – a major step for the future of more privacy-focused AI. With a 3-billion parameter option available, it’s possible Apple may be quietly preparing something every bit as capable as Google Gemini, albeit behind the scenes.

In February, Apple’s CEO Tim Cook revealed that AI features are planned for future mac releases, but did not give further details, and has so far not released general purpose models for commercial use.

This has led to intense speculation that Apple are considering heading in a slightly different direction to the rest of big tech – perhaps keen to differentiate themselves in key areas.

However, Apple have already launched machine learning frameworks that hint at AI models running on M-series chipsets, an image editing model, and a UI-based model that would allow an AI to navigate a smartphone or tablet.

All this leaves Apple fans waiting until at least the Apple Worldwide Developers Conference in June, or other annual announcements during September, to see what’s waiting behind the scenes.

For IT Support and expertise, please contact our team today.

Microsoft Copilot will release to users worldwide from 26th September 2023.

The flagship natural language AI tool that previewed back in March of this year brings ChatGPT functions to a whole range of Microsoft productions including your favourite Microsoft 365 apps, Azure, Bing and into Windows itself.

At a colourful launch event in New York, Microsoft CEO Satya Nadella debuted the official release of Copilot and demonstrated ways AI can revolutionise everyday tasks – whether that be writing emails, interpreting or generating images, transcribing handwriting into maths, or intelligently answering questions.

Copilot assistance aims to make Microsoft 365 apps including Teams, Word, Excel and more increasingly powerful – with users able to simply request complex data handling tasks rather than manipulating the data manually themselves. For more creative work, Copilot can also generate visual results within longstanding Microsoft apps like Paint.

This presents some fascinating opportunities for companies using the Microsoft 365 suite – common tasks such as ‘Re-write this email more professionally’, ‘Summarise this meeting’ and ‘Make me a set of PowerPoint slides based on this document’ would all be achievable within a few seconds.

Features from the Bing public preview have also made it into the release version – with users able to choose ‘More Creative’ ‘More Balanced’ and ‘More Precise’ language options to give the AI’s output a different tone, and draw on the wider web-based dataset that is previously unknown to the user.

For example, in another impressive example of Copilot using web and user data intelligently, an American user uploads a photo of a UK plug adaptor and asks ‘Will this work in London?’ Copilot is not only able to check based on an understanding of the image, but understands what the user means by ‘work’ and is able to confirm that the plug is the correct choice.

Some interesting guard rails have also been rolled in – Microsoft have previously stated that a firebreak exists between user data and the web dataset used to train the AI, but Copilot can use both to respond to prompts. Images created using the next version of DALL.E will also be crytographically signed as ‘Created by AI’, effectively signing the content as AI-generated.

Copilot will soon begin appearing to most users on the Windows toolbar – coinciding with the new upgrade of Windows 11 that is due on 26th September. Within Microsoft 365, Copilot Chat functions will begin appearing in app updates for Enterprise licence customers from 1st November 2023.

For Microsoft expertise and support, please contact our team today.

Zoom has introduced revisions to its Terms of Service that have sparked significant controversy over user privacy.

It appears that Zoom may begin using calls between millions of users around the world to train AI products. The updated terms contain two standout sections, 10.2 and 10.4, which hold considerable implications for the extent to which Zoom can leverage user data. These segments specify Zoom’s entitlements to gather and utilise “Service Generated Data,” encompassing telemetry data, product usage information, diagnostic data, and analogous content collected in conjunction with users’ use of Zoom’s services or software.

Explicitly specified in Zoom’s revised policy is its exclusive ownership over Service Generated Data. This dominion extends to the company’s prerogatives to alter, disseminate, process, exchange, retain, and warehouse said data “for any purpose, within the limits and parameters stipulated by applicable law.”

“…You agree to grant and hereby grant Zoom a perpetual, worldwide, non-exclusive, royalty-free, sublicensable, and transferable license and all other rights required or necessary to redistribute, publish, import, access, use, store, transmit, review, disclose, preserve, extract, modify, reproduce, share, use, display, copy, distribute, translate, transcribe, create derivative works, and process Customer Content and to perform all acts with respect to the Customer Content, including AI and ML training and testing.”

Of particular concern is the overt declaration of the company’s right to employ this data for the training and enhancement of machine learning and artificial intelligence systems, with the inclusion of algorithm and model refinement. This particular clause raises the question of a lack of opt-out alternatives, something that is bound to fuel intense debate about user-generated consent and individual privacy.

Zoom justifies these measures as indispensable for delivering services to patrons, supporting the services, and enhancing its range of offerings including software and other products. Nevertheless, the implications inherent in these clauses are all-encompassing, particularly as they seemingly permit Zoom to harness customer data for any purpose associated with the activities or processes detailed in section 10.3.

Remarkably, Zoom has not issued any comments regarding these amendments yet. While Zoom’s motivations may center on elevating their platform and providing an improved experience, the extent and profundity of these adjustments are likely to cause disquiet among numerous users, particularly security and privacy advocates, thereby prompting inquiries into how their data is being leveraged.

For more software expertise and support, please contact our team today

Microsoft has announced a new Teams experience for Windows, replacing the existing version with a faster, cleaner Teams app.

The new version seeks to dramatically improve user experience, introduces powerful-AI based tools to more users, and addresses performance issues that have crept into Teams since the platform’s astonishing success during the Covid-19 pandemic – which saw rapid adoption by organisations around the world and an explosion of new features.

Microsoft estimates the new communication and collaboration app is now twice as fast, but uses approximately 50% less system memory, completing initial loading in under 10 seconds compared to over 20 seconds in ‘Classic’ Teams in independent benchmarked testing.

Part of that extra speed – in loading, scrolling and searching – has been achieved by tidying many of the extra controls that have been added to teams, simplifying the design for a ‘cleaner’ viewing window. Users who need access to multiple accounts will also find it easier to switch between work environments.

Microsoft are also highlighting the extensive integration of ChatGPT AI tools in the form of Copilot – which debuted for Microsoft 365 a few weeks ago. Within Teams, AI assistance will be able to summarise meetings, write chat or documents on your behalf, and answering questions mid-discussion flow.

While we’re not quite convinced about 3D avatars (!) the new Teams can highlight when individuals were speaking, automatically break long meetings into key ‘chapters’ for video review, redesigns Teams channels to feel more like the gathering point they’re intended to be, and use AI mid-chat in a way that has untold potential.

Overall the new design is more visually impressive, with customisation options that are available in public preview now, and available to Microsoft 365 users later in 2023.

For Microsoft 365 Expertise and support, please contact our team today.

Microsoft has announced a GPT-4 powered AI assistant for the Microsoft 365 Office apps named ‘Copilot’, in an impressive array of preview videos.

Copilot promises to bring OpenAI’s groundbreaking natural language assistant into Word, Excel, Outlook, Powerpoint, Teams and more – helping users to write documents automatically with prompts, analyse spreadsheets and more.

Copilot can also use other Office documents as its source material, allowing powerful cross referencing commands such as ‘Make a Presentation based on my notes in Document.docx’, or copy styles based on examples of your other work.

The new tool, announced at Microsoft’s ‘Future of Work’ event in March, follows a string of recent Chat GPT related developments for the tech giant – which Microsoft promises to dramatically cut the time required to produce first-drafts, as well as rapidly cross-reference with data the user doesn’t have to hand.

Within Microsoft Teams, Copilot promises to summarise meetings based on text analysis of the transcript, or even recap for late attendees based on what they have missed.

Unlike the original ChatGPT version 3 preview, Copilot promises to implement some of the advances used in the trial version of Bing – with references and links embedded inside the AI’s responses demonstrating why it chose that answer.

Don’t expect to see Chat GPT 4 doing all your work for you quite yet though – Copilot remains in a trial phase with twenty select Microsoft customers, with the preview set to expand in coming months. (No Clippy, I wasn’t trying to write a letter – Copilot will do it for me.)

For Microsoft 365 expertise and support, please contact our team today.

Microsoft have released an early preview of AI-powered search integrated into Bing.

Available via Bing.com, the new tool is the first sign of Microsoft’s multibillion dollar investment in artificial intelligence company OpenAI – and uses the same technology first seen in the open preview of ChatGPT 3.5. AI ‘Copilot’ now not only delivers search results, but can generates longer text-based responses to questions, and form responses to more abstract queries that don’t have an obvious web destination.

The move comes just 48 hours after Google’s rival AI chatbot, Bard, generated a factual error in a promotional video, wiping an estimated $100bn off Google’s total market value.

ChatGPT’s ‘chat prompt’ format also allows the user to respond to Bing in turn, and continue the conversation to get a refined response without having to restart their search from scratch.

In each case Bing’s responses contain automatically embedded links that allow the user to follow-up the origins of each recommendation, even where the citations originate from a wide range of sources.

A new sidebar in Microsoft Edge takes this one step further by allowing Bing to ‘read’ each web page you visit, and intelligently re-format the results – picking out key details or presenting data back in a new format.

Sample queries and a waiting list are already available via Bing.com, but the preview is expected to be made widely available to millions of end users within weeks as Microsoft bets big on the future of search.

For IT support and technical expertise, please contact our team today.

AI lent a helping hand to one of our technical support teams last week to help Lineal save a local business from an email hack.

At 07:40 GMT on a seemingly normal week day, Barracuda Sentinel issued an alert to Lineal to say an account had been accessed from a suspicious location.  It seemed a malicious actor, appearing to be from Nigeria, compromised one of a client’s finance department email accounts, and created a forwarding/delete rule in the inbox.

Barracuda Sentinel’s AI email protection caught the account takeover attempt, and as a result, we were able to mitigate and resolve a significant threat to one of our customers. Barracuda Sentinel detects both account takeover attempts and attacks launched from compromised accounts.

Corporate account takeover presents a significant new threat to business. Hackers gain access to email accounts and use them as tools to launch subsequent targeted attacks, internally and against external targets – who themselves fall victim.

Account takeover or attacks that originate from these accounts are almost impossible to detect as they don’t use the usual impersonation techniques—they come from a legitimate account and appear to be from a trusted source, allowing the attacker to initiate sophisticated financial scams.

Lineal automatically picked up the alert & create an incident in Barracuda Sentinel.  Sentinel remediated the issue with an immediate password reset, disconnecting all active logon sessions for the user and deletion of any rules created during the incident time.  Within 40 minutes this potentially disastrous event was avoided.