With o3 having reached AGI, OpenAI turns its sights toward superintelligence
One of the challenges of achieving AGI is defining it. As of yet, researchers
and the broader industry do not have a concrete description of what it will be
and what it will be able to do. The general consensus, though, is that AGI
will possess human-level intelligence, be autonomous, have self-understanding,
and will be able to “reason” and perform tasks that it was not trained to do.
... Going beyond AGI, “superintelligence” is generally understood to be AI
systems that far surpass human intelligence. “With superintelligence, we can
do anything else,” Altman wrote. “Superintelligent tools could massively
accelerate scientific discovery and innovation well beyond what we are capable
of doing on our own.” He added, “this sounds like science fiction right now,
and somewhat crazy to even talk about it.” However, “we’re pretty confident
that in the next few years, everyone will see what we see,” he said,
emphasizing the need to act “with great care” while still maximizing benefit.
... OpenAI set out to build AGI from its founding in 2015, when the concept of
AGI, as Altman put it to Bloomberg, was “nonmainstream.” “We wanted to figure
out how to build it and make it broadly beneficial,” he wrote in his blog
post.
Bridging the execution gap – why AI is the new frontier for corporate strategy
Imagine a future where leadership teams are not constrained by outdated
processes but empowered by intelligent systems. In this world, CEOs use AI to
visualise their entire organisation’s alignment, ensuring every department
contributes to strategic goals. Middle managers leverage real-time insights to
adapt plans dynamically, while employees understand how their work drives the
company’s mission forward. Such an environment fosters resilience, innovation,
and engagement. By turning strategy into a living, breathing entity,
organisations can adapt to challenges and seize opportunities faster than ever
before. The road to this future is not without challenges. Leaders must
embrace cultural change, invest in the right technologies, and commit to
continuous learning. But the rewards – a thriving, agile organisation capable
of navigating the complexities of the modern business landscape – are well
worth the effort. The execution gap has plagued organisations for decades, but
the tools to overcome it are now within reach. AI is more than a technological
advancement; it is the key to unlocking the full potential of corporate
strategy. By embracing adaptability and leveraging AI’s transformative
capabilities, businesses can ensure their strategies do not just survive but
thrive in the face of change.
Google maps the future of AI agents: Five lessons for businesses
Google argues that AI agents represent a fundamental departure from
traditional language models. While models like GPT-4o or Google’s Gemini excel
at generating single-turn responses, they are limited to what they’ve learned
from their training data. AI agents, by contrast, are designed to interact
with external systems, learn from real-time data and execute multi-step tasks.
“Knowledge [in traditional models] is limited to what is available in their
training data,” the paper notes. “Agents extend this knowledge through the
connection with external systems via tools.” This difference is not just
theoretical. Imagine a traditional language model tasked with recommending a
travel itinerary. ... At the heart of an AI agent’s capabilities is its
cognitive architecture, which Google describes as a framework for reasoning,
planning and decision-making. This architecture, known as the orchestration
layer, allows agents to process information in cycles, incorporating new data
to refine their actions and decisions. Google compares this process to a chef
preparing a meal in a busy kitchen. The chef gathers ingredients, considers
the customer’s preferences and adapts the recipe as needed based on feedback
or ingredient availability. Similarly, an AI agent gathers data, reasons about
its next steps and adjusts its actions to achieve a specific goal.
AI agents will change work forever. Here's how to embrace that transformation
The business world is full of orthodoxies, beliefs that no one questions
because they are thought to be "just the way things are". One such orthodoxy
is the phrase: "Our people are the difference". A simple Google search can
attest to its popularity. Some companies use this orthodoxy as their official
or unofficial tagline, a tribute to their employees that they hope sends the
right message internally and externally. They hope their employees feel
special and customers take this orthodoxy as proof of their human goodness.
Other firms use this orthodoxy as part of their explanation of what makes
their company different. It's part of their corporate story. It sounds nice,
caring, and positive. The only problem is that this orthodoxy is not true. ...
Another way to put this is that individual employees are not fixed assets.
They do not behave the same way in all conditions. In most cases, employees
are adaptable and can absorb and respond to change. The environment,
conditions, and potential for relationships cause this capacity to express
itself. So, on the one hand, one company's employees are the same as any other
company's employees in the same industry. They move from company to company,
read the same magazines, attend similar conventions, and learn the same
strategies and processes.
Gen AI is transforming the cyber threat landscape by democratizing vulnerability hunting
Identifying potential vulnerabilities is one thing, but writing exploit code
that works against them requires a more advanced understanding of security
flaws, programming, and the defense mechanisms that exist on the targeted
platforms. ... This is one area where LLMs could make a significant impact:
bridging the knowledge gap between junior bug hunters and experienced exploit
writers. Even generating new variations of existing exploits to bypass
detection signatures in firewalls and intrusion prevention systems is a
notable development, as many organizations don’t deploy available security
patches immediately, instead relying on their security vendors to add
detection for known exploits until their patching cycle catches up. ... “AI
tools can help less experienced individuals create more sophisticated exploits
and obfuscations of their payloads, which aids in bypassing security
mechanisms, or providing detailed guidance for exploiting specific
vulnerabilities,” Nițescu said. “This, indeed, lowers the entry barrier within
the cybersecurity field. At the same time, it can also assist experienced
exploit developers by suggesting improvements to existing code, identifying
novel attack vectors, or even automating parts of the exploit chain. This
could lead to more efficient and effective zero-day exploits.”
GDD: Generative Driven Design
The independent and unidirectional relationship between agentic platform/tool
and codebase that defines the Doctor-Patient strategy is also the greatest
limiting factor of this strategy, and the severity of this limitation has
begun to present itself as a dead end. Two years of agentic tool use in the
software development space have surfaced antipatterns that are increasingly
recognizable as “bot rot” — indications of poorly applied and problematic
generated code. Bot rot stems from agentic tools’ inability to account for,
and interact with, the macro architectural design of a project. These tools
pepper prompts with lines of context from semantically similar code snippets,
which are utterly useless in conveying architecture without a high-level
abstraction. Just as a chatbot can manifest a sensible paragraph in a new
mystery novel but is unable to thread accurate clues as to “who did it”,
isolated code generations pepper the codebase with duplicated business logic
and cluttered namespaces. With each generation, bot rot reduces RAG
effectiveness and increases the need for human intervention. Because bot
rotted code requires a greater cognitive load to modify, developers tend to
double down on agentic assistance when working with it, and in turn rapidly
accelerate additional bot rotting.
Someone needs to make AI easy
Few developers did a better job of figuring out how to effectively use AI than
Simon Willison. In his article “Things we learned about LLMs in 2024,” he
simultaneously susses out how much happened in 2024 and why it’s confusing.
For example, we’re all told to aggressively use genAI or risk falling behind,
but we’re awash in AI-generated “slop” that no one really wants to read. He
also points out that LLMs, although marketed as the easy path to AI riches for
all who master them, are actually “chainsaws disguised as kitchen knives.” He
explains that “they look deceptively simple to use … but in reality you need a
huge depth of both understanding and experience to make the most of them and
avoid their many pitfalls.” If anything, this quagmire got worse in 2024.
Incredibly smart people are building incredibly sophisticated systems that
leave most developers incredibly frustrated by how to use them
effectively. ... Some of this stems from the inability to trust AI to
deliver consistent results, but much of it derives from the fact that we keep
loading developers up with AI primitives (similar to cloud primitives like
storage, networking, and compute) that force them to do the heavy lifting of
turning those foundational building blocks into applications.
Making the most of cryptography, now and in the future
The mathematicians and cryptographers who have worked on these NIST
algorithms expect them to last a long time. Thousands of people have
already tried to poke holes into them and haven’t yet made any meaningful
progress toward defeating them. So, they are “probably” OK for the time being.
But as much as we would like to, we cannot mathematically rule out that they
cannot be broken. This means that for commercial enterprises looking to
migrate to new cryptography, they should be braced to change again and again —
whether that is in five years, 10 years, or 50 years. ... Up until now most
cryptography was mostly implicit and not under direct control of the
management. Putting more controls around cryptography would not only safeguard
data today, but it would provide the foundation to make the next transition
easier. ... Cryptography is full of single points of failure. Even if
your algorithm is bulletproof, you might end up with a faulty implementation.
Agility helps us move away from these single points of failure, allowing us to
adapt quickly if an algorithm is compromised. It is therefore crucial for
CISOs to start thinking about agility and redundancy.
Data 2025 outlook: AI drives a renaissance of data
Though not all the technology building blocks are in place, many already are.
Using AI to crawl and enrich metadata? Automatically generate data pipelines?
Using regression analysis to flag data and model drift? Using entity extraction
to flag personally identifiable information or summarize the content of
structured or unstructured data? Applying machine learning to automate data
quality resolution and data classification? Applying knowledge graphs to RAG?
You get the idea. There are a few technology gaps that we expect will be
addressed in 2025, including automating the correlation between data and model
lineage, assessing the utility and provenance of unstructured data, and
simplifying generation of vector embeddings. We expect in the coming year that
bridging data file and model lineage will become commonplace with AI governance
tools and services. And we’ll likely look to emerging approaches such as data
observability to transform data quality practices from reactive to proactive.
Let’s start with governance. In the data world, this is hardly a new discipline.
Though data governance over the years has drawn more lip service than practice,
for structured data, the underlying technologies for managing data quality,
privacy, security and compliance are arguably more established than for
AI.
Beware the Rise of the Autonomous Cyber Attacker
Research has already shown that teams of AIs working together can find and
exploit zero-day vulnerabilities. A team at the University of Illinois
Urbana-Champaign created a “task force” of AI agents that worked as a supervised
unit and effectively exploited vulnerabilities they had no prior knowledge of.
In a recent report, OpenAI also cited three threat actors that used ChatGPT to
discover vulnerabilities, research targets, write and debug malware and setup
command and control infrastructure. The company said the activity offered these
groups “limited, incremental (new) capabilities” to carry out malicious cyber
tasks. ... “Darker” AI use
has, in part, prompted many of today’s top thinkers to support regulations. This
year, OpenAI CEO Sam Altman said: “I’m not interested in the killer robots
walking on the street … things going wrong. I’m much more interested in the very
subtle societal misalignments, where we just have these systems out in society
and through no particular ill intention, things go horribly wrong.” ...
Theoretically, regulation may reduce unintended or dangerous use among
legitimate users, but I’m certain that the criminal economy will appropriate
this technology. As CISOs deploy AI more broadly, attackers’ abilities will
concurrently soar.
Quote for the day:
"Leadership is a dynamic process that
expresses our skill, our aspirations, and our essence as human beings." --
Catherine Robinson-Walker
