Quote for the day:
"A leader is one who knows the way, goes the way, and shows the way." -- John C. Maxwell
Synthetic data and the risk of ‘model collapse’
There is a danger of an ‘ouroboros’ here, or a snake eating its own tail. Models
can be ‘poisoned’ with data that is passed on in addition to malicious prompts.
While usually caused by sabotage, this can also be unintentional: AI models
sometimes hallucinate, including when they are generating data for their LLM
descendant. With enough ongoing errors, a new LLM risks performing worse than
its predecessors. At its core, it’s a simple case of garbage in, garbage out.
The logical end state is a total ‘model collapse‘, where drivel overtakes
anything factual and makes an LLM dysfunctional. Should this happen (and it may
have happened with GPT-4.5), AI model makers are forced to pull back to an
earlier checkpoint, reassess their data or be forced to make architectural
changes. ... In short, a high degree of expertise is required for each step in
the AI process. Currently, attention is focused on the initial building of the
foundation models on the one hand and the actual implementation of GenAI on the
other. The importance of training data was touched upon in 2023 because online
organizations regularly felt robbed. In essence: it made headlines, which is why
we all became aware of the intricacies of training data. Now that the flow of
online retrievable data is ending, AI players are grasping for an alternative
that is creating new problems.
Automated Workflow Perfection Is a Job in Itself
“The fragmented nature of automation – spanning robotic process automation,
business process management, workflow tools and AI-powered solutions all further
complicates consistent measurement,” lamented Gaudette. “Market segment overlap
presents another challenge. As technologies increasingly converge, traditional
category boundaries blur. A document processing solution might be classified
under workflow automation by one analyst and digital process automation by
another, creating inconsistent market size calculations.” Other survey
“findings” from Custom Workflows’ analysis report suggest that the integration
of artificial intelligence with traditional automation represents a particularly
powerful growth catalyst. McKinsey’s own analysis reveals that while basic
automation delivers 20-30% cost reductions, intelligent automation incorporating
AI can achieve 50-70% savings while simultaneously improving quality and
customer experience. ... As the market for workflow automation now goes into
what we might call an amplified state of flux, it appears that current
automation adoption follows a classic bell curve distribution, with most
organizations clustered in the middle stages of implementation maturity.
Surprisingly, smaller organizations often outperform their larger counterparts
when it comes to automation success.
The hidden risk in SaaS: Why companies need a digital identity exit strategy
To reduce dependency on external SaaS providers, organizations should consider
taking back control of their digital identity infrastructure. This doesn’t mean
abandoning cloud services altogether, but rather strategically deploying
identity management solutions that provide ownership and portability.
Self-hosted identity solutions running on private cloud or on-premises
environments can offer greater control. Businesses should also consider
multi-cloud identity architectures allowing authentication and access control to
function across different cloud providers. ... Organizations must closely
monitor data sovereignty laws and adjust their infrastructure accordingly.
Ensuring that identity solutions comply with shifting regulations will help
avoid legal and operational risks. To avoid being caught off guard, it’s
important for IT teams to understand what’s going on behind the scenes rather
than entirely outsourcing their infrastructure. For the highest level of
preparedness, organizations can manage identity infrastructure systems
themselves, reducing reliance on third party SaaS companies for critical
functions. If teams understand the inner workings of their identity management,
they will be better placed to develop an emergency response plan with predefined
steps to transition services in case of sudden geopolitical changes.
Why Your Business Needs an AI Innovation Unit
An AI innovation unit should always support sustainable and strategic
organizational growth through the ethical and impactful application and
integration of AI, McDonagh-Smith says. "Achieving this mission involves
identifying and deploying AI technologies to solve complex and simple business
problems, improving efficiency, cultivating innovation, and creating measurable
new organizational value." A successful unit, McDonagh-Smith states, prioritizes
aligning AI initiatives with the enterprise's long-term vision, ensuring
transparency, fairness, and accountability in its AI applications. ... An AI
innovation unit leader is foremost a business leader and visionary, responsible
for helping the enterprise embrace and effectively use AI in an ethical and
responsible manner, Hall says. "The leader needs to understand the risk and
concerns, but also AI governance and frameworks." He adds that the leader should
also be realistic and inspiring, with an understanding of the hype curve and the
technology's potential. ... An AI innovation unit requires a collaborative
culture that bridges silos within the organization and commits to continuous
reflection and learning, McDonagh-Smith says. "The unit needs to establish
practical partnerships with academic institutions, tech startups, and AI thought
leadership groups to create flows of innovation, intelligence, and business
insights."
How to avoid the AI complexity trap
When done right, AI enables simplicity, cutting across layers of complexity --
but with limits. "AI is not a silver bullet," said Richard Demeny, a software
development consultant, formerly with Arm. "LLMs under the hood actually use
probabilities, not understanding, to give answers. It's humans who design,
build, and implement systems, and while AI may automate some entry-level roles
and certainly bring significant productivity gains, it cannot replace the amount
of practical experience IT decision-makers need to make the right trade-offs."
... To keep both AI and IT complexity at bay, "deployment of AI needs to be
thoughtful," said Hashim. "Focus on the simplicity of user experience, quality
of AI, and its ability to get things done," she said. "Uplevel all your
employees with AI so that your organization as a whole can be more productive
and happy." Consistency is the key to managing complexity, Howard said.
Platforms, for example, "make things consistent. So you're able to do things --
sometimes very complicated things -- in consistent ways and standard ways that
everybody knows how to use them. Even something as simple as definitions or
taxonomy. If everybody is speaking the same language, so a simplified taxonomy,
then it's much easier to communicate."
Outsmart the skills gap crisis and build a team without recruitment
Team augmentation involves engaging external software engineers from a partner
company to complement an existing in-house team. This approach provides
companies with the flexibility to quickly scale their technical resources up or
down, depending on the project’s needs, and plug any capability gaps inside
their teams. It can be crucial to the success of businesses whose product is
software, or relies on software, as it enables businesses to scale their team
and projects flexibly without the risks involved with growing an in-house team.
... It allows companies to access a diverse range of skills and expertise that
may not be available in-house. Companies can quickly ramp up their technical
resources and tackle projects that require specialised skills or knowledge
whilst onboarding engineers that can bring fresh ideas and perspectives to the
project. Having access to this expertise quickly is often of paramount
importance as companies compete to grow. For instance, if a company needs to
design, develop, and support a mobile app, but its in-house team lacks the
necessary skills and experience, it can quickly engage a team of engineers who
specialise in mobile app development to work on the project. This approach can
help companies save time and resources and ensure that their projects are
completed on time and to a high standard.
Taking AI Commoditization Seriously
Commoditization is the process of products or services becoming “standardized,
marketable objects.” Any given unit of a commodity, from corn to crude oil, is
generally interchangeable with and sells for the same price as others.
Commoditization of frontier models could emerge in a few ways. Perhaps, as Yann
LeCun predicts, open-source models could equal or surpass closed-source
performance. Or perhaps competing firms continue finding ways to match each
other’s developments. Such competition has more above-board variants—top-tier
engineers at different firms keeping pace with each other—and less. Consider,
for instance, OpenAI’s allegations against DeepSeek of inappropriate copying.
... The emergence of new, decentralized AI threat vectors could offer the powers
that be a common enemy. This might present a unique opportunity for US-China
collaboration. Modern US-China collaboration has required tangible mutual
interest to succeed. The most famous modern US-China agreement, the
Nixon/Kissinger-Mao/Zhou normalization of US-China relations, occurred in large
part to overcome a perceived common threat in the USSR. When few companies
control cutting-edge frontier models, preventing third-party model misuse is
comparatively simple. Fewer frontier developers imply fewer sites to monitor for
malicious actors.
Making Architecturally Significant Decisions
Architectural decisions are at the root of our practice but they are often hard
to spot. The vast majority of decisions get processed at the team level and do
not apply architectural thinking or have an architect involved at all. This
approach can be a benefit in agile organizations if managed and communicated
effectively. ... Envision an enterprise or company, then imagine all the teams
in the organization working in parallel on changes, remember to add in
maintenance teams and operations teams doing ‘keep the lights running’ work. ...
To effectively manage decisions, the architecture team should put in place a
decision management process early in its lifecycle, by making critical
investments into how the organization is going to process decision point in the
architecture engagement model. During the engagement methodology update and the
engagement principles definition, the team will decide what levels of decisions
must be exposed in the repository and their limits in duration, quality and
effort. These principles will guide the decision methods for the entire team
until the next methodology update. There are numerous decision methods and
theories in the marketplace in making better decisions. The goal of the
architecture decision repository is to ensure that decisions are made clearly,
with appropriate tools and with respect for traceability.
What is predictive analytics? Transforming data into future insights
Predictive analytics draws its power from many methods and technologies,
including big data, data mining, statistical modeling, ML, and assorted
mathematical processes. Organizations use predictive analytics to sift through
current and historical data to detect trends, and forecast events and conditions
that should occur at a specific time, based on supplied parameters. With
predictive analytics, organizations can find and exploit patterns contained
within data in order to detect risks and opportunities. Models can be designed,
for instance, to discover relationships between various behavior factors. Such
models enable the assessment of either the promise or risk presented by a
particular set of conditions, guiding informed decision making across various
categories of supply chain and procurement events. ... Predictive analytics
makes looking into the future more accurate and reliable than previous tools. As
such it can help adopters find ways to save and earn money. Retailers often use
predictive models to forecast inventory requirements, manage shipping schedules,
and configure store layouts to maximize sales. Airlines frequently use
predictive analytics to set ticket prices reflecting past travel
trends.
C-Suite Leaders Must Rewire Businesses for True AI Value
AI's true value doesn't come from incremental gains but emerges when workflows
are transformed completely. McKinsey found 21% of companies using gen AI have
redesigned workflows and seen significant effect on their bottom-line. Morgan
Stanley redesigned client interactions by integrating AI-powered assistants.
Rather than just automating document retrieval, the company embedded AI into
workflows, enabling advisers to generate customized reports and insights in real
time. This improved efficiency and enhanced customer experience through more
data-driven, personalized interactions. Boston Consulting Group highlighted that
companies embedding AI into core business workflows report 40% higher process
efficiency and 25% faster output. For CIOs and AI leaders, this highlights a
crucial point. Deploying AI without rethinking workflows resembles putting a
turbo engine in a low-end car. The real competitive advantage comes from
integrating AI into the fabric of business operations and not in standalone
tasks. ... AI is becoming a core function that enhances decision-making,
automates tasks and drives innovation. McKinsey's report emphasized that AI's
biggest value lies in large-scale transformation, not isolated use cases.
