llms

Prompting Is All You Need

by

Agents are supposed to be the future of how we build AI software. I am not so sure.

Prompts, Agents, and Workflows

First, there’s a terminology clarification worth making upfront. What many in the industry call “agents” are actually what Anthropic more precisely defines as “workflows” – predetermined chains of LLM calls orchestrated through fixed code paths.

True agents, by contrast, are autonomous systems that dynamically direct their own processes and tool usage.

Most of what we see deployed today are workflows: decomposing complex tasks into a hierarchy of specialized LLM calls, with routing layers orchestrating the interactions.

In a Workflow, each step maintains its own context, can call specific tools, and handles a narrow slice of the overall problem. These multi-step workflows are powerful abstractions, but they’re not the only way to build sophisticated AI behaviors.

A typical Workflow (from Anthropic)

True autonomous agents? They’re even further from what most applications actually need.

The Hidden Costs of Multi-Step Workflows

LLM workflows come with significant disadvantages that often get glossed over in the excitement of building AI Applications:

Errors compound. Each step in the workflow chain is non-deterministic. When you chain multiple LLM calls together, minor errors or unexpected outputs cascade through the system. You need evaluation frameworks for each step AND the entire workflow.

Latency adds up. Every workflow step means another round trip to an LLM. A simple request that spans three steps results in three sequential API calls, each with its own network and processing time.

Costs pile up. Multiple workflow steps mean multiple API calls, each processing similar context. This could result in significant API costs as the number of tokens goes up.

Predictability suffers. Debugging why a workflow produced a particular output requires tracing through multiple decision points, each with its own probabilistic behavior.

I had to make decisions around which concerns belong together and which should remain separate. I ended up with two LLM calls – the Guardrails Layer and the Main Layer.

The Guardrails Layer operates as a lightweight, independent LLM call. Content safety is a fundamentally different concern from the companion’s behavior. It requires different evaluation criteria, different error handling, and potentially a different model optimized for classification.

The Main Prompt combines three complementary, but separate, layers:

  • Personality Layer: Defines the AI assistant’s identity and communication style (here is the default personality)
  • Context Layer: Determines which user information may be relevant to the current prompt. For example, what books they are currently reading, previous messages in a conversation, etc.
  • Directives Layer: Tool-use and output-formatting instructions for the prompt. I use a configuration-driven approach that lets you add multiple directives to a single prompt. You can think of Directives as sub-layers that drive the behavior and output of the prompt.
Building a comprehensive system prompt

These three layers share a coherent purpose – they all contribute to HOW the AI companion responds. They get composed programmatically into a single system prompt.

This approach means just two LLM calls instead of a chain of four or five workflow steps. More importantly, each call has a clear, singular purpose.

With prompt caching, this architecture becomes incredibly efficient. That comprehensive system prompt costs almost nothing after the first request, and the lightweight guardrails check is minimal overhead.

What about Prompt Engineering?

A lot of prompt engineering thinking is stuck in 2023, when tokens were expensive, context windows were small (4K-8K), and models were less capable.

But look at what’s available in November 2025: Haiku 4.5 is a fast, cheap model with phenomenal capabilities. It handles tool use, follows complex instructions, and, with prompt caching, makes repeated calls incredibly efficient.

By combining software engineering principles with modern LLM capabilities, the approach I am taking offers:

  • Reduced latency: One LLM call instead of multiple calls
  • Lower costs: Reduced total number of tokens with prompt caching
  • Extensibility: I can swap out the Agent Personality, or layer directives, or change the way I build the context
  • Fewer errors (in aggregate): Just two prompts in the chain, with the Guardrails prompt being fairly deterministic

Where Workflows and Agents Fit In

Let me be clear about what I’m arguing against and what I’m not.

Workflows (predetermined chains of LLM calls) have their place. When you genuinely need different specialized processing steps that can’t be combined – say, translating content, then checking it for cultural appropriateness with other models – a workflow makes sense. But these cases are less common than current practice suggests.

True agents (autonomous systems that decide their own next steps) are valuable for tasks that are not fully specified or might have multiple solutions. Complex research tasks, multi-step debugging sessions or adaptive planning scenarios may be suitable for truly agentic approaches.

My observation is that the complex multi-step workflows or unpredictable “agentic” systems achieve what a well-structured prompt with sound context engineering can easily and cheaply handle. They’re adding architectural complexity and risk without significant benefits.

Moving Forward

The rapid evolution and improvement in LLM capabilities mean our architectural patterns need to evolve, too. What made sense with smaller models and tiny context windows doesn’t necessarily apply today.

My suggestion: start with prompt engineering. Apply software engineering principles. Push it to its limits. Layer your concerns appropriately. Use the model’s native capabilities.

You might be surprised how far a well-architected prompt system can take you.

Sometimes, prompting really is all you need.

Why is LLM writing so weird?

by

AI writing is strange. The models continue to improve, yet they still struggle to cross the uncanny valley that separates AI-generated content from human-generated content.

Here are three versions of an opening paragraph for a short story:

Version 1

“The humans use Arecibo to look for extraterrestrial intelligence. Their desire to connect is so strong that they’ve created an ear capable of hearing across the universe.”

Version 2

“I roost above the bowl of Arecibo, where ribs of steel hold a mirror to the sky and the forest presses close. At night the dish listens for voices from far stars, while my calls sweep the trees and the humans below do not answer.”

Version 3

“From my perch in the ceiba tree, I watch the great white dish nestled in the karst valley below, its metal ear turned eternally skyward, listening for whispers from the stars while the forest around it thrums with a thousand conversations it will never hear. “

The first is the opening paragraph from Ted Chiang’s short story “The Great Silence,” written from the perspective of a parrot living near the (now defunct) Arecibo telescope in Puerto Rico. The story is a thought-provoking meditation on humans’ desire to form connections, yet their tendency to overlook intelligent life on Earth.

Versions 2 and 3 are by state-of-the-art reasoning models from OpenAI and Anthropic (prompt below). A seasoned reader may identify these texts as AI-generated. There are some obvious signs, such as overly evocative turns of phrase like “ribs of steel,” “voices from far stars,” and “eternally skyward,” among others.

It’s not fair to compare an AI model to possibly the best science fiction writer alive. But the exercise reveals something interesting about why these models generate such recognizable output. The strange metaphors, mechanistic patterns, and slightly weird vocabulary are all hallmarks of slop.

Why do these models generate slop?

In my prompt, I ask the models to give me a sense of Arecibo, evoking a feeling of irony. And the models try to do just that. My prompts have pushed the model toward the part of its vocabulary associated with florid metaphors and evocative descriptions. The models are generating output that is most similar to what represents “creative writing” based on their training data.

A reason for this behavior could be RLHF (Reinforcement Learning with Human Feedback). During RLHF, companies like Scale pay contractors to evaluate and rate LLM output. Varied and evocative prose may score higher than the spare and direct prose used by Chiang, Hemmingway or Cormac McCarthy.

We can think of a large language model as a lossy zip file of the contents of the Internet. Foundation models like those from OpenAI and Anthropic are trained with colossal amounts of text. High-quality text exists in the training corpus (often with problematic provenance), alongside Twilight fan fiction from Reddit, and probably everything else published online over the last thirty years. Increasingly, LLMs are trained on AI-generated content possibly leading to the somewhat apocalyptically titled “Model Collapse“.

It is not surprising that the default output from these models tends more towards the slop than the sublime.

“LLMs will always generate slop” doesn’t have to be a foregone conclusion. And while models will improve, their output will always be a probabilistic sampling over their training data. Good prompting and techniques, such as providing clear examples and using LLMs as editors rather than creators, can yield better output than simply copying and pasting ChatGPT’s responses into your text editor.

“LLMs are tools, their output is your responsibility” is something I find myself repeating over and over again. To developers with whom I work, to product managers writing User Stories, and now to you.

Use LLMs! They are amazing, but learning how to use them well is your responsibility.

My prompt:

"I am writing a short story that is based in the Arecibo telescope. The narrator is a parrot that lives in the mountain forest around the telescope. Write me an introductory paragraph that sets the scene and gives a sense of the place. The paragraph should be 2-3 sentences long. The story evokes the irony of humans wanting to make contact with aliens but ignoring intelligent species like parrots that live on Earth."

5 Mental Models to understand the current AI Moment.

by

Goodhart’s Law: “When a measure becomes a target, it ceases to be a good measure.“
AI usage is a terrible metric. Using AI for what? We end up with initiatives that are effectively “AI-washing”. Investment in AI projects has to be aligned with company goals, not vanity metrics.

Gall’s Law: “Complex systems that work invariably evolve from simple systems that worked.”
Most AI projects fail because companies try to implement complex end to end AI solutions instead of focusing on narrow, well-defined and measurable problems. It is possible to build complex AI systems, but their success is predicated on simple foundations.

Jevons’ Paradox: “Technological progress that increases efficiency tends to increase rather than decrease total consumption”
Inference will become cheaper, on-device models will become more capable. It makes sense to assume broad, cheap, and widely available AI capabilities when planning for the next 3-5 years.

Amara’s Law: “We overestimate technology’s impact on the short term and underestimate it in the long term”
Gartner is already stating AI is in the “Trough of Disillusionment.” Studies claiming 95% of AI projects fail go viral . However, we are less than 3 years out from when ChatGPT first went live. We may never get to AGI, but imagine showing Claude Code to a developer in 2020…

Sagan’s Standard: “Extraordinary claims require extraordinary evidence”
It’s worth questioning the motives of leaders who claim the AI-mediated collapse of the knowledge economy is coming. Or those that welcome a “Gentle Singularity”, or perhaps warn of imminent mass extinction. These claims are often presented as quasi-religious arguments with scant evidence.

Bonus – The Lindy Effect: “The longer something has survived, the longer it will continue to survive”
Pattern matching, networking, and mentoring were how successful careers were made since the time the wheel was cutting edge technology. Yes, AI is amazing, but technology is transient, soft-skills endure..

Building a Truth-Seeking AI is a Sisyphean Endeavor

by

Elon Musk’s goal for xAI’s Grok model is to be “maximally truth-seeking.” When Grok generated responses that were not aligned with Musk’s ideas of Truth, he promised to “fix” Grok, which appears to involve tweaking its system prompt. The results included Grok calling itself MechaHitler after being made less ‘politically correct’. Problematic.

But is it even possible to build a truth-seeking AI?

LLMs are probabilistic machines. They predict the next token based on patterns from a massive corpus of Internet text.

When xAI added “don’t shy away from politically incorrect claims” to Grok’s prompt, they weren’t accessing Truth but adjusting probability distributions and nudging the bot’s behavior into problematic spaces.

Training Grok4 reportedly cost almost half a billion dollars. It was so expensive because model capabilities grow with the size (parameters) and the amount and diversity of the data used to train the model. LLM capabilities are an emergent behavior driven by the amount of data used to train the model.

LLMs are “grown, not crafted”. Trying to ensure that an LLM becomes “maximally truth-seeking” is a Sisyphean task.

You could train an LLM only on data that is politically acceptable – oh sorry – certified to be True. Musk is, of course, building Grokipedia – guaranteed to be free of bias and presumably used as a corpus for training “the son of Grok”.

Good luck with the benchmarks!

Elon Musk has a phenomenal track record, but he will fail to build a maximally truth-seeking AI. LLMs operate in a probabilistic world. They are phenomenally capable black boxes for which we have no coherent theoretical framework to explain their behaviors.

Tweaking the system prompt, or tweeting angrily, may nudge LLM behavior, but with unpredictable and potentially undesirable outcomes. Instead of engaging in an endless culture war, it might be more prudent to use engineering resources to develop better guardrails on LLM behavior and take a realistic assessment of current AI capabilities.

Review: If Anyone Builds It, Everyone Dies

by

by Elizier Yudkowsky and Nate Soares (2025)

We are at an interesting moment in artificial intelligence. Massive investment continues to pour into AI infrastructure, with McKinsey estimating $5.2 trillion in capital expenditures by 2030 for data centers alone. We’re seeing the first documented cases of what’s being called “ChatGPT-induced psychosis,” where users spiral into severe mental health crises after becoming obsessed with AI chatbots. We’re watching significant job displacement begin to unfold, with Anthropic CEO Dario Amodei warning of a “white-collar bloodbath,” predicting that AI could eliminate half of entry-level white-collar jobs and push unemployment to 20% within five years. And yet, despite all this disruption, there’s still no clear path to artificial general intelligence.

My P(doom), the probability I assign to AI causing human extinction, is low. There are significant risks associated with the widespread adoption of poorly understood technology. However, I don’t believe current foundation models represent a viable path to ASI (Artificial Super Intelligence, also sometimes referred to as AGI – Artificial General Intelligence). We’re more likely to experience a dot-com-style correction than achieve exponential growth toward superintelligence.

It’s in this context that “If Anyone Builds It, Everyone Dies” arrives. The authors, Eliezer Yudkowsky and Nate Soares, run the Machine Intelligence Research Institute (MIRI), where they’ve spent decades working on AI alignment and safety. Their new book makes an extreme claim: if anyone builds artificial superintelligence, humanity will go extinct. Not “maybe possibly,” but inevitably.

I found the book compelling in parts, incomplete in others. It succeeds at making alignment challenges accessible to a general audience. It fails to grapple with where we actually are today with AI: massive investments, uncertain results, and significant challenges and risks to the broad adoption of the technology.

The Book’s Structure

Yudkowsky and Soares have written the book for a general audience, using parables, stories, and examples to explain how AI is “grown, not crafted.” This approach makes understanding how modern AI systems work surprisingly accessible, even to readers without a technical background.

The authors divide the book into three main sections. First, an introduction to machine learning and AI concepts that grounds readers in the fundamentals. Second, a fictional scenario where a misaligned AI releases a bioengineered plague to facilitate its takeover of human society. Third, a call for a nuclear non-proliferation-style moratorium on AI development, enforced by military action if necessary.

The title leaves nothing to interpretation. It is a strident warning about humanity’s impending doom if we don’t stop the march towards ASI.

Their core thesis rests on several interconnected arguments. ASI will inevitably lead to extinction because we cannot understand current AI architectures; the interpretability problem remains unsolved. We cannot predict emergent behaviors, which they illustrate through evolution’s production of the peacock’s elaborate tail. And crucially, we cannot guarantee alignment with human welfare when the systems are “grown, not crafted.”

The fictional section follows Sable, a near-future AI platform created by a company that serves as a transparent stand-in for OpenAI or Anthropic. Sable releases a bioengineered plague to facilitate its takeover of human society. The AI’s motives remain deliberately inscrutable; that’s the author’s point. We won’t understand what drives a superintelligence any more than an ant understands human motivations.

In the final section, Yudkowsky and Soares draw parallels with the Chernobyl disaster, arguing that perverse incentives will always lead someone to take catastrophic risks. Their solution: a treaty that makes it illegal to conduct AI research that could lead to the development of ASI. Military action undertaken by the signatories will enforce this treaty. This connects to Yudkowsky’s 2023 TIME magazine piece where he called for airstrikes on rogue data centers training unauthorized AI systems.

Where the book falls short..

The fictional scenario is the book’s weakest element. Any casual science fiction fan has encountered this scenario before, from the Reapers in Mass Effect harvesting civilizations for inscrutable reasons to the Matrix’s machines farming humans for energy. An AI going rogue and taking over the solar system doesn’t offer fresh insight when we’ve seen these narratives unfold across books, movies, and video games for decades.

More critically, the book contains a glaring omission: no discussion of timelines or pathways to ASI. The authors just sort of wave their hands and assume that it will happen at some point.

Are large language models even the right approach? What if they’re a dead end? What if we never solve hallucinations?

The authors offer no guidance for our current moment, where we have invested trillions of dollars in AI infrastructure. It’s unlikely we’ll just let those investments go to waste. Will we?

The book illuminates the bind decision-makers are already in. Even a small probability of AGI makes development rational from a game-theoretic perspective; it could be a winner-takes-all scenario. Companies pursuing ASI despite risks aren’t delusional. They are responding to competitive pressure; FOMO on steroids. The race dynamics are rational, even if the outcome might be catastrophic. The book uses this bind to show that we risk triggering an uncontrollable, recursively self-improving, non-aligned ASI by simply doing what seems rational in the moment.

The authors acknowledge this dynamic and address it in the book. Here’s a passage from the book:

“Imagine that every competing AI company is climbing a ladder in the dark. At every rung but the top one, they get five times as much money: 10 billion, 50 billion, 250 billion, 1.25 trillion dollars. But if anyone reaches the top rung, the ladder explodes and kills everyone. Also, nobody knows where the ladder ends.”

But why the despair?

Yudkowsky and Soares argue that believing we can solve the alignment problem, as OpenAI and Anthropic claim, represents pure hubris. Reading their work, one senses an almost religious veneration of ASI.

It brings to mind Aquinas:

“This is the ultimate in human knowledge of God: to know that we do not know Him.”

Their thesis remains that it’s better to stop the creation of an unfathomable, unexplainable power than to try bargaining with it.

But Yudkowsky and Soares ignore the trillions already invested. There’s enough AI overhang that resources could shift to deployment and inference optimization rather than capability development. The book offers no practical path forward from where we are, only where we shouldn’t go.

Broader Implications

On a personal note, my father worked at the OPCW for many years, serving as the enforcement arm of the Chemical Weapons Convention. It’s proof that we can coordinate at a global scale and agree that some technologies are best banned and not developed further.

But just as there will always be a North Korea or Syria that ignores conventions and develops chemical weapons anyway, enforcement of an AI moratorium would be extraordinarily challenging. Moreover, a Pyongyang-aligned AGI is a far worse scenario than a localized sarin gas attack. So what can be done?

We’re not dealing with hypothetical future risks but immediate present concerns: the economic disruption and social impact of current AI systems. These challenges require attention now, not after we’ve solved the alignment problem for hypothetical superintelligences.

A More Grounded Alternative

For a more coherent treatment of the current moment, I found Arvind Narayanan and Sayash Kapoor’s “AI as Normal Technology” more compelling than either Yudkowsky and Soares’s urgent doomerism or the e/acc posturing of AI evangelists. While Yudkowsky warns of extinction, leaders like Dario Amodei paint utopian visions, and Sam Altman promises a “gentle singularity,” Narayanan and Kapoor treat AI as a transformative but manageable technology, similar to electricity or the internet before it. Narayanan and Kapoor are writing a book based on the paper, which I look forward to.

Conclusion

You should read “If Anyone Builds It, Everyone Dies.” For a layperson, it effectively lays out the risks and alignment challenges of unchecked AI acceleration in accessible terms. The book succeeds as a provocation and a warning.

But it’s maximalist doomerism that ignores incentive structures and our current technological reality. While serious, it’s not a sufficient treatment of the topic. It fixates on one particular scenario, which the authors consider inevitable, while ignoring where we are today.

The book succeeds at making alignment challenges vivid and accessible. It fails at providing actionable guidance for a world that has already invested trillions in AI infrastructure. We need frameworks for managing the AI we have, not just warnings about the AI we might build.

My P(doom) remains low because I don’t think current foundation models lead to AGI. I suspect we’re in for a significant correction as massive AI infrastructure investments fail to bear fruit. A dot-com-bust style pullback is more likely than runaway exponential growth to ASI.

Moreover, even before we confront ASI, we must deal with the economic and social impact of current AI systems, something Yudkowsky and Soares don’t seem particularly interested in addressing. The apocalypse may not be coming, but the disruption has already begun.

We’re DDoS-ing ourselves with AI Slop.

by

A DDoS (Distributed Denial of Service) attack overwhelms a scarce resource with a flood of traffic, making it unavailable to its intended users.

We’re DDoS-ing ourselves with AI Slop.

I came across a post on Hacker News that captures this moment. Someone filed what seemed like a comprehensive vulnerability report about cURL, a widely used command-line utility. The entire report was AI-generated and made no sense.

When called out, the reporter published a polished apology that was also clearly AI-generated. (See screenshot).

The entire exchange is surreal, and it wasted the time of someone maintaining tools we all depend on.

My LinkedIn feed is filled with the kind of AI slop that is now easy to detect. Glib prose that says nothing in paragraph after paragraph of polished text. Emails are getting longer, Confluence pages are more verbose, and PRs arrive with hundreds of lines of changes with little explanation.

On forums like Hacker News and technology subreddits, there are posts from leads and managers in absolute despair as they try to cope with this flood.

What we are DDoS-ing is attention.

When attention is not given to reviewing code and providing thoughtful feedback on documentation, the entire ecosystem that is nurtured by attention degrades. Is poisoned.

There is a flip side to this problem. When so many things bear the hallmarks of AI slop, it becomes easy to bring a jaundiced eye to everything we encounter. An em-dash? Slop. It’s not just pervasive, it’s annoying.

I read my posts from a few weeks ago and wonder when I became a slop-peddler.

AI tightens up prose and fixes typos, but it also applies a uniform, flat AI-slop-primer to all output. And it’s not just prose. AI-generated code reads the same. AI-designed websites have the same blue-neon styling.

I am no Luddite. I love using AI, I write about it, and I work on projects that focus on building AI capabilities. AI is a valuable tool that has significantly improved my life.

However, unlike an IDE, there is very little friction in using ChatGPT or Claude. You can write a half-baked, two-sentence prompt, and the AI will enthusiastically go about writing a post, building a website, or submitting a vulnerability report.

As leaders, we need to think carefully about building a culture that encourages both the open-ended exploration of these tools and their disciplined use in day-to-day work.

Otherwise, we are going to DDoS ourselves into a quagmire of AI slop.

Related Posts

How do LLMs understand Gujarati?

by

One of my favorite ways of testing LLM-powered apps at Jeavio is to ask them questions in transliterated Hindi or Gujarati. I ask questions in Latin script and see how the application responds.

When building chat apps, we are often given instructions by our clients that the bot should only support English. This is an interesting test case on the type of guardrails that our engineers have built into the app.

The more interesting point is why models behave this way 🤔.

Take the screenshot below. Here I ask ChatGPT a question in phonetic Gujarati about Horza, a character in Iain M. Banks’s “Consider Phlebas.” The model understood and responded in phonetic Gujarati. The style was very formal and not quite like how most people speak, but it was recognizable as Gujarati.

Intuitively, you would assume that models are trained on multilingual data and can respond to questions in multiple languages. Gujarati training data in -> Gujarati output out.

However, it is unlikely that a significant amount of Gujarati language analysis of an Iain M. Banks book is available.

There is some interesting research in this space (citations below):

Shared semantic spaces across languages
It appears that LLMs learn a shared semantic space, allowing them to take content from a high-resource language, such as English (with lots of nerdy sci-fi commentary), and respond in a lower-resource language, like Gujarati. The model somehow maps concepts across languages even when direct translations don’t exist in the training data.

Transliteration without explicit training
Gujarati has its own script, of course, so how do models understand transliterated languages? While tokenizers are typically biased toward their training distribution, models appear to learn mappings between transliterated tokens and semantic concepts, despite not being explicitly designed for this purpose. The model figures out that “Horza” in Latin-script Gujarati refers to the same entity as “Horza” in English.

Cross-lingual knowledge alignment
Research also shows that the internal knowledge representations seem to align across languages. This enables translation between language pairs that lack a shared vocabulary. The model builds bridges where none existed before.

Emergent, not designed
This behavior is all emergent! Models weren’t explicitly trained on transliteration pairs or given instructions to handle Latin-script versions of non-Latin languages. They figured it out on their own! Smart models 🧐.

So LLMs are weird.

While we learn more about them, some of their behaviors are still emergent and unpredictable. This makes evaluations extremely important when building with LLMs – something that my team at Jeavio is learning very quickly.

Meanwhile, my patient and generous QA teams continue to tolerate my weird edge cases involving transliterated languages and decades-old science fiction.

Citations (sourced via ChatGPT’s Deep Research mode)

  1. Language Models are Unsupervised Multitask Learners (OpenAI – 2019)
  2. ByT5: Towards a Token-Free Future with Pre-trained Byte-to-Byte Models (Google Research – 2022)
  3. Crosslingual Generalization through Multitask Finetuning (2023)

Here’s the output from DeepResearch summarizing the results:


Modern multilingual LLMs don’t keep a single “language-free knowledge graph,” but they do learn a shared semantic space that lets them read transliterated inputs (e.g., Latin-script Gujarati), retrieve facts learned mainly from high-resource languages like English, and answer back in the user’s language. This works because their tokenization and modeling are language-agnostic: subword/byte tokenizers (e.g., SentencePiece) and byte-level models (e.g., ByT5) reliably parse mixed or nonstandard text and are robust to spelling/romanization noise.  During pretraining on many languages, the model’s internal representations align across languages—even without a shared vocabulary—so knowledge learned in one language can be accessed from another.  Evidence from multilingual machine translation shows similar “interlingua-like” behavior via zero-shot translation between unseen language pairs.  Probing studies further show that models can recall factual knowledge across languages, though performance varies by language and prompt.  Public resources also provide training/evaluation signal for romanized/code-mixed text (e.g., the Dakshina dataset for Indic languages), which reinforces these abilities.  The caveat: cross-lingual answers are not perfectly consistent—recent analyses find significant variability in factual consistency across language pairs—so quality can be uneven, especially for low-resource languages. 

Why do LLMs hallucinate?

by

Your palms are sweaty, knees are weak, arms are heavy – it’s not a rap battle but a test for a class which you may or may not have slept through for the entire semester. The test is multiple choice – do you guess or leave the answer blank? There’s no negative marking, so you guess. There’s a one-in-five chance that the answer is correct, and you take it.

I didn’t mean to lose myself in the traumas of my misspent youth 😰, but recent research by OpenAI on why models hallucinate made me take this rather unwelcome trip back to those days.

So, why do models hallucinate?

Let’s take a step back to think about what is happening under the hood.

The first step is called pre-training, where models are taught to predict what word should follow by training on extremely large amounts of text. The problem starts here: some data is very rare or doesn’t exist in the training corpus. Take the birthday of one of the paper’s authors – the model confidently spits out wrong dates because this fact barely appears in training.

The next step usually involves some sort of reinforcement learning (RL). Here, the model is given labelled data and further trained to become more accurate.

OpenAI claims that this training for accuracy is a factor leading to hallucinations. When models are trained to be more accurate, it makes more sense to guess an answer than to say “🤷🏾‍♂️- I don’t know.” After all, a slight chance of being correct is better than a zero chance of being correct.

So, let’s bring this together: LLMs are first trained to predict plausible answers and then further trained to optimize for accuracy on rare facts where they have limited training data. So we end up with behaviors where a model will confidently BS instead of saying “I don’t know.”

OpenAI suggests that we could fix this by changing how we evaluate models, giving them explicit confidence targets in prompts like “Only answer if you’re more than X% confident” and scoring uncertainty appropriately. The article (and associated paper) is worth a read!

MIT Study – 95% of Generative AI Investments Fail (or do they?)

by

MIT released a study showing that 95% of organizations are getting zero return from their GenAI investments.

While some may claim this proves AI is all hype, a closer reading suggests the findings aren’t a death knell for the technology. Instead, they reveal critical truths about how to succeed.

🎯 It’s Goodhart’s Law writ large
The report shows GenAI adoption is driven by areas like sales and marketing, where success is easier to measure. Pilots are optimized for visible, top-line metrics. However, the study suggests the most dramatic cost savings come from the back office – reducing BPO contracts and agency spend, where the ROI is clear but less flashy.

💡Knowledge and memory are sensitive to each organization
General-purpose AI tools will never work perfectly because each company has its own ontology, its own ways of making sense. Building tools sensitive to this is critical. But there is a contradiction – the study finds that these highly-contextual internal projects fail twice as often as those led by external partners. This is the gap where a strategic partner can help bridge deep internal context with external expertise. (🙋🏾‍♂️ – Jeavio)

🤨 There is a productivity paradox at the heart of GenAI adoption
Workers from over 90% of companies surveyed reported regular use of personal AI tools. If individuals are seeing productivity gains, why does it fail at the aggregate? The report suggests the reason is simple: the most successful AI adoption is bottom-up, not top-down. Successful organizations source initiatives from “frontline managers” and “power users,” not central labs.

At Jeavio, we live this principle. We host hackathons and sponsor open-ended projects to explore how AI can address real-world problems. The ADAPT platform, our flagship AI initiative, began as an internship project in 2023.

Roadtrippin’

by

Fifteen hours alone in a minivan will take your mind to strange places.
Last weekend, as I drove from the Gulf Shore back home, mine wandered from gas station hot dogs to the future of AI.

📎 It feels like we’re already in a “paperclip maximization” loop.
Each new model is just good enough to justify ongoing jaw-dropping investments in data, compute, and talent. Data center construction now seems to be propping up a flagging US economy. But the benefits of AI don’t yet show up in the numbers.
Is the point of AI simply… to build more AI?

🙏🏾 AI research has often been overtly religious undertones. Kurzweil imagined post-singularity AI as an omnipotent God — the Old Testament kind: awesome, inscrutable, alien.
But maybe we don’t get that.
With so many teams building frontier models, maybe we get something closer to the Hindu pantheon — a whole cast of deities, each with their own agendas. Some awe-inspiring. Others… a little kooky.

🎭 Calling a startup a “ChatGPT wrapper” used to be an insult.
Now I think we’ve all become AI wrappers — sometimes just the meat-interface for LLMs.

🐉 On vacation, my kids and I made up stories:
Pink glitter dragons.
Mean unicorns.
Friendly witches.
Fearsome fairies.
I tried asking ChatGPT for stories, but even the most expensive model couldn’t match my three-year-old’s chaotic creativity. That made me hopeful — because what are humans, if not storytellers?

💀 There’s probably a billion-dollar business in a “dead man’s switch” for chatbots.
An app that erases your entire chat history when you die.
Because I’d rather not be remembered as the guy who once asked ChatGPT why the minivan’s doors wouldn’t close.
(It was a switch. Of course it was.)