Quote for the day:
"Accept responsibility for your life. Know that it is you who will get you where you want to go, no one else." -- Les Brown
AI That Thinks Like Us: New Model Predicts Human Decisions With Startling Accuracy
“We’ve created a tool that allows us to predict human behavior in any
situation described in natural language – like a virtual laboratory,” says
Marcel Binz, who is also the study’s lead author. Potential applications range
from analyzing classic psychological experiments to simulating individual
decision-making processes in clinical contexts – for example, in depression or
anxiety disorders. The model opens up new perspectives in health research in
particular – for example, by helping us understand how people with different
psychological conditions make decisions. ... “We’re just getting started and
already seeing enormous potential,” says institute director Eric Schulz.
Ensuring that such systems remain transparent and controllable is key, Binz
adds – for example, by using open, locally hosted models that safeguard full
data sovereignty. ... The researchers are convinced: “These models have
the potential to fundamentally deepen our understanding of human cognition –
provided we use them responsibly.” That this research is taking place at
Helmholtz Munich rather than in the development departments of major tech
companies is no coincidence. “We combine AI research with psychological theory
– and with a clear ethical commitment,” says Binz. “In a public research
environment, we have the freedom to pursue fundamental cognitive questions
that are often not the focus in industry.”
Collaboration is Key: How to Make Threat Intelligence Work for Your Organization
A challenge with joining threat intelligence sharing communities is that a lot
of threat information is generated and needs to be shared daily. For already
resource-stretched teams, it can be extra work to pull together, share a threat
intelligence report, and filter through the incredible volumes of information.
Particularly for smaller organizations, it can be a bit like drinking from a
firehose. In this context, an advanced threat intelligence platform (TIP) can be
invaluable. A TIP has the capabilities to collect, filter, and prioritize data,
helping security teams to cut through the noise and act on threat intelligence
faster. TIPs can also enrich the data with additional contexts, such as threat
actor TTPs (tactics, techniques and procedures), indicators of compromise
(IOCs), and potential impact, making it easier to understand and respond to
threats. Furthermore, an advanced TIP can have the capability to automatically
generate threat intelligence reports, ready to be securely shared within the
organization’s threat intelligence sharing community Secure threat intelligence
sharing reduces risk, accelerates response and builds resilience across entire
ecosystems. If you’re not already part of a trusted intelligence-sharing
community, it is time to join. And if you are, do contribute your own valuable
threat information. In cybersecurity, we’re only as strong as our weakest link
and our most silent partner.
Google study shows LLMs abandon correct answers under pressure, threatening multi-turn AI systems
The researchers first examined how the visibility of the LLM’s own answer
affected its tendency to change its answer. They observed that when the model
could see its initial answer, it showed a reduced tendency to switch, compared
to when the answer was hidden. This finding points to a specific cognitive bias.
As the paper notes, “This effect – the tendency to stick with one’s initial
choice to a greater extent when that choice was visible (as opposed to hidden)
during the contemplation of final choice – is closely related to a phenomenon
described in the study of human decision making, a choice-supportive bias.” ...
“This finding demonstrates that the answering LLM appropriately integrates the
direction of advice to modulate its change of mind rate,” the researchers write.
However, they also discovered that the model is overly sensitive to contrary
information and performs too large of a confidence update as a result. ...
Fortunately, as the study also shows, we can manipulate an LLM’s memory to
mitigate these unwanted biases in ways that are not possible with humans.
Developers building multi-turn conversational agents can implement strategies to
manage the AI’s context. For example, a long conversation can be periodically
summarized, with key facts and decisions presented neutrally and stripped of
which agent made which choice.
Building stronger engineering teams with aligned autonomy
Autonomy in the absence of organizational alignment can cause teams to drift in different directions, build redundant or conflicting systems, or optimize for local success at the cost of overall coherence. Large organizations with multiple engineering teams can be especially prone to these kinds of dysfunction. The promise of aligned autonomy is that it resolves this tension. It offers “freedom within a framework,” where engineers understand the why behind their work but have the space to figure out the how. Aligned autonomy builds trust, reduces friction, and accelerates delivery by shifting control from a top-down approach to a shared, mission-driven one. ... For engineering teams, their north star might be tied to business outcomes, such as enabling a frictionless customer onboarding experience, reducing infrastructure costs by 30%, or achieving 99.9% system uptime. ... Autonomy without feedback is a blindfolded sprint, and just as likely to end in disaster. Feedback loops create connections between independent team actions and organizational learning. They allow teams to evaluate whether their decisions are having the intended impact and to course-correct when needed. ... In an aligned autonomy model, teams should have the freedom to choose their own path — as long as everyone’s moving in the same direction.How To Build a Software Factory
Of the three components, process automation is likely to present the biggest
hurdle. Many organizations are happy to implement continuous integration and
stop there, but IT leaders should strive to go further, Reitzig says. One
example is automating underlying infrastructure configuration. If developers
don’t have to set up testing or production environments before deploying code,
they get a lot of time back and don’t need to wait for resources to become
available. Another is improving security. Though there’s value in continuous
integration automatically checking in, reviewing and integrating code, stopping
there can introduce vulnerabilities. “This is a system for moving defects into
production faster, because configuration and testing are still done manually,”
Reitzig says. “It takes too long, it’s error-prone, and the rework is a tax on
productivity.” ... While the software factory standardizes much of the
development process, it’s not monolithic. “You need different factories to
segregate domains, regulations, geographic regions and the culture of what’s
acceptable where,” Yates says. However, even within domains, software can serve
vastly different purposes. For instance, human resources might seek to develop
applications that approve timesheets or security clearances. Managing many
software factories can pose challenges, and organizations would be wise to
identify redundancies, Reitzig says.
Why Scaling Makes Microservices Testing Exponentially Harder
You’ve got clean service boundaries and focused test suites, and each team can
move independently. Testing a payment service? Spin up the service, mock the
user service and you’re done. Simple. This early success creates a reasonable
assumption that testing complexity will scale proportionally with the number of
services and developers. After all, if each service can be tested in isolation
and you’re growing your engineering team alongside your services, why wouldn’t
the testing effort scale linearly? ... Mocking strategies that work beautifully
at a small scale become maintenance disasters at a large scale. One API change
can require updating dozens of mocks across different codebases, owned by
different teams. ... Perhaps the most painful scaling challenge is what happens
to shared staging environments. With a few services, staging works reasonably
well. Multiple teams can coordinate deployments, and when something breaks, the
culprit is usually obvious. But as you add services and teams, staging becomes
either a traffic jam or a free-for-all — and both are disastrous. ... The teams
that successfully scale microservices testing have figured out how to break this
exponential curve. They’ve moved away from trying to duplicate production
environments for testing and are instead focused on creating isolated slices of
their production-like environment.India’s Digital Infrastructure Is Going Global. What Kind of Power Is It Building?
India’s digital transformation is often celebrated as a story of frugal
innovation. DPI systems have allowed hundreds of millions to access ID, receive
payments, and connect to state services. In a country of immense scale and
complexity, this is an achievement. But these systems do more than deliver
services; they configure how the state sees its citizens: through biometric
records, financial transactions, health databases, and algorithmic scoring
systems. ... India’s digital infrastructure is not only reshaping domestic
governance, but is being actively exported abroad. From vaccine certification
platforms in Sri Lanka and the Philippines to biometric identity systems in
Ethiopia, elements of India Stack are being adopted across Asia and Africa. The
Modular Open Source Identity Platform (MOSIP), developed in Bangalore, is now in
use in more than twenty countries. Indeed, India is positioning itself as a
provider of public infrastructure for the Global South, offering a postcolonial
alternative to both Silicon Valley’s corporate-led ecosystems and China’s
surveillance-oriented platforms. ... It would be a mistake to reduce India’s
digital governance model to either a triumph of innovation or a tool of
authoritarian control. The reality is more of a fragmented and improvisational
technopolitics. These platforms operate across a range of sectors and are shaped
by diverse actors including bureaucrats, NGOs, software engineers, and civil
society activists.
Chris Wright: AI needs model, accelerator, and cloud flexibility
As the model ecosystem has exploded, platform providers face new complexity. Red Hat notes that only a few years ago, there were limited AI models available under open user-friendly licenses. Most access was limited to major cloud platforms offering GPT-like models. Today, the situation has changed dramatically. “There’s a pretty good set of models that are either open source or have licenses that make them usable by users”, Wright explains. But supporting such diversity introduces engineering challenges. Different models require different model customization and inference optimizations, and platforms must balance performance with flexibility. ... The new inference capabilities, delivered with the launch of Red Hat AI Inference Server, enhance Red Hat’s broader AI vision. This spans multiple offerings: Red Hat OpenShift AI, Red Hat Enterprise Linux AI, and the aforementioned Red Hat AI Inference Server under the Red Hat AI umbrella. Along the are embedded AI capabilities across Red Hat’s hybrid cloud offerings with Red Hat Lightspeed. These are not simply single products but a portfolio that Red Hat can evolve based on customer and market demands. This modular approach allows enterprises to build, deploy, and maintain models based on their unique use case, across their infrastructure. This from edge deployments to centralized cloud inference, while maintaining consistency in management and operations.Data Protection vs. Cyber Resilience: Mastering Both in a Complex IT Landscape
Traditional disaster recovery (DR) approaches designed for catastrophic events
and natural disasters are still necessary today, but companies must implement a
more security-event-oriented approach on top of that. Legacy approaches to
disaster recovery are insufficient in an environment that is rife with
cyberthreats as these approaches focus on infrastructure, neglecting
application-level dependencies and validation processes. Further, threat actors
have moved beyond interrupting services and now target data to poison, encrypt
or exfiltrate it. As such, cyber resilience needs more than a focus on recovery.
It requires the ability to recover with data integrity intact and prevent the
same vulnerabilities that caused the incident in the first place. ... Failover
plans, which are common in disaster recovery, focus on restarting Virtual
Machines (VMs) sequentially but lack comprehensive validation.
Application-centric recovery runbooks, however, provide a step-by-step approach
to help teams manage and operate technology infrastructure, applications and
services. This is key to validating whether each service, dataset and dependency
works correctly in a staged and sequenced approach. This is essential as
businesses typically rely on numerous critical applications, requiring a more
detailed and validated recovery process.
Rethinking Distributed Computing for the AI Era
The problem becomes acute when we examine memory access patterns. Traditional
distributed computing assumes computation can be co-located with data,
minimizing network traffic—a principle that has guided system design since the
early days of cluster computing. But transformer architectures require
frequent synchronization of gradient updates across massive parameter
spaces—sometimes hundreds of billions of parameters. The resulting communication
overhead can dominate total training time, explaining why adding more GPUs often
yields diminishing returns rather than the linear scaling expected from
well-designed distributed systems. ... The most promising approaches involve
cross-layer optimization, which traditional systems avoid when maintaining
abstraction boundaries. For instance, modern GPUs support mixed-precision
computation, but distributed systems rarely exploit this capability
intelligently. Gradient updates might not require the same precision as forward
passes, suggesting opportunities for precision-aware communication protocols
that could reduce bandwidth requirements by 50% or more. ... These architectures
often have non-uniform memory hierarchies and specialized interconnects that
don’t map cleanly onto traditional distributed computing abstractions.
/filters:no_upscale()/articles/agentic-ai-architecture-framework/en/resources/96figure-2-1752043185139.jpg)
