Connect with us
DAPA Banner

Crypto World

How to Reduce Non-Determinism and Hallucinations in Large Language Models (LLMs)

Published

on

How to Reduce Non-Determinism and Hallucinations in Large Language Models (LLMs)

In recent months, two separate pieces of research have shed light on two of the most pressing issues in large language models (LLMs): their
non-deterministic nature and their tendency to
hallucinate. Both phenomena have a direct impact on the
reliability,
reproducibility, and
practical usefulness of these technologies.

On the one hand,
Thinking Machines, led by former OpenAI CTO Mira Murati, has published a paper proposing ways to make LLMs return the
exact same answer to the
exact same prompt every time, effectively defeating non-determinism. On the other hand,
OpenAI has released research identifying the root cause of hallucinations and suggesting how they could be significantly reduced.

Let’s break down both findings and why they matter for the future of AI.

The problem of non-determinism in LLMs

Anyone who has used ChatGPT, Claude, or Gemini will have noticed that when you type in the exact same question multiple times, you don’t always get the same response. This is what’s known as
non-determinism: the same input does not consistently lead to the same output.

Advertisement

In some areas, such as creative writing, this variability can actually be a feature; it helps generate fresh ideas. But in domains where
consistency, auditability, and reproducibility are critical — such as healthcare, education, or scientific research — it becomes a serious limitation.

Why does non-determinism happen?

The most common explanation so far has been a mix of two technical issues:

  1. Floating-point numbers: computer systems round decimal numbers, which can introduce tiny variations.
  2. Concurrent execution on GPUs: calculations are performed in parallel, and the order in which they finish can vary, changing the result.

However, Thinking Machines argues that this doesn’t tell the whole story. According to their research, the real culprit is batch size.

When a model processes multiple prompts at once, it groups them into batches (or “carpools”). If the system is busy, the batch is large; if it’s quiet, the batch is small. These variations in batch size subtly change the order of operations inside the model, which can ultimately influence which word is predicted next. In other words, tiny shifts in the order of addition can completely alter the final response.

Thinking Machines’ solution

The key, they suggest, is to keep internal processes consistent regardless of batch size. Their paper outlines three core fixes:

Advertisement
  1. Batch-invariant kernels: ensure operations are processed in the same order, even at the cost of some speed.
  2. Consistent mixing: use one stable method of combining operations, independent of workload.
  3. Ordered attention: slice input text uniformly so the attention mechanism processes sequences in the same order each time.

The results are striking: in an experiment with the Qwen 235B model, applying these methods produced 1,000 identical completions to the same prompt, rather than dozens of unique variations.

This matters because determinism makes it possible to audit, debug, and above all, trust model outputs. It also enables stable benchmarks and easier verification, paving the way for reliable applications in mission-critical fields.


The problem of hallucinations in LLMs

The second major limitation of today’s LLMs is hallucination: confidently producing false or misleading answers. For example, inventing a historical date or attributing a theory to the wrong scientist.

Why do models hallucinate?

According to OpenAI’s paper, hallucinations aren’t simply bugs; they are baked into the way we train LLMs. There are two key phases where this happens:

  1. Pre-training: even with a flawless dataset (which is impossible), the objective of predicting the next word naturally produces errors. Generating the
    right answer is harder than checking whether an answer
    is right.
  2. Post-training (reinforcement learning): models are fine-tuned to be more “helpful” and “decisive”. But current metrics reward correct answers while penalising both mistakes
    and admissions of ignorance. The result? Models learn that it’s better to bluff with a confident but wrong answer than to say “I don’t know”.

This is much like a student taking a multiple-choice exam: leaving a question blank guarantees zero, while guessing gives at least a chance of scoring. LLMs are currently trained with the same incentive structure.

OpenAI’s solution: behavioural calibration

The proposed solution is surprisingly simple yet powerful: teach models when not to answer. Instead of forcing a response to every question, set a confidence threshold.

Advertisement
  • If the model is, for instance, more than 75% confident, it answers.
  • If not, it responds:
    “I don’t know.”

This technique is known as behavioural calibration. It aligns the model’s stated confidence with its actual accuracy.

Crucially, this requires rethinking benchmarks. Today’s most popular evaluations only score right and wrong answers. OpenAI suggests a three-tier scoring system:

  • +1 for a correct answer
  • 0 for “I don’t know”
  • –1 for an incorrect answer

This way, honesty is rewarded and overconfident hallucinations are discouraged.

Signs of progress

Some early users report that GPT-5 already shows signs of this approach: instead of fabricating answers, it sometimes replies,
“I don’t know, and I can’t reliably find out.” Even Elon Musk praised this behaviour as an impressive step forward.

The change may seem small, but it has profound implications: a model that admits uncertainty is far more trustworthy than one that invents details.


Two sides of the same coin: reliability and trust

What makes these two breakthroughs especially interesting is how complementary they are:

Advertisement
  • Thinking Machines is tackling
    non-determinism, making outputs consistent and reproducible.
  • OpenAI is addressing
    hallucinations, making outputs more honest and trustworthy.

Together, they target the biggest barrier to wider LLM adoption: confidence. If users — whether researchers, doctors, teachers, or policymakers — can trust that an LLM will both give reproducible answers and know when to admit ignorance, the technology can be deployed with far greater safety.


Conclusion

Large language models have transformed how we work, research, and communicate. But for them to move beyond experimentation and novelty, they need more than just raw power or creativity: they need trustworthiness.

Thinking Machines has shown that non-determinism is not inevitable; with the right adjustments, models can behave consistently. OpenAI has demonstrated that hallucinations are not just random flaws but the direct result of how we train and evaluate models, and that they can be mitigated with behavioural calibration.

Taken together, these advances point towards a future of AI that is more transparent, reproducible, and reliable. If implemented at scale, they could usher in a new era where LLMs become dependable partners in science, education, law, and beyond.

Source link

Advertisement
Continue Reading
Click to comment

You must be logged in to post a comment Login

Leave a Reply

Crypto World

SEC Seeks Public Comment on Crypto Handling in OTC Broker-Dealer Rule

Published

on

Crypto Breaking News

The US Securities and Exchange Commission is moving to reduce years of ambiguity around a broker-dealer reporting rule that had limited which assets could be quoted on the over-the-counter (OTC) market. Rule 15c2-11, originally adopted in 1971 to curb penny-stock fraud, requires broker-dealers to keep current public information about a listed issuer before publishing quotes. In 2021, the rule was reinterpreted to also cover fixed-income securities, a shift that drew backlash from market participants and raised questions about crypto securities. In a Monday statement, the SEC proposed an amendment to limit the rule’s scope to equity securities, effectively reversing the 2021 interpretation. The move arrives amid a broader regulatory push to clarify how crypto assets fit within traditional market structures.

Hester Peirce, a commissioner who leads the SEC’s crypto task force, welcomed the proposal and argued that the commission had created years of uncertainty through a 2020 amendment and its 2021 application. She noted that, by the letter of Rule 15c2-11, the rule has always applied to quotations of a “security,” but market participants and observers understood it to cover only OTC equity securities. The commissioner stressed that long-term relief should have been granted while the agency assessed whether extending the rule to fixed income was appropriate and amended the rule as needed. Instead, she said, the commission issued several rounds of limited relief—often lasting only a few months—fostering ongoing uncertainty in the market.”

Key takeaways

  • The SEC proposes narrowing Rule 15c2-11’s reporting obligations to equity securities on OTC markets, reversing the 2021 interpretation that extended it to fixed-income assets.
  • The agency has opened a 60-day public comment period to gather feedback on how “equity securities” should be defined and whether crypto assets might fall under that category.
  • The proposal highlights the commission’s intent to reduce regulatory ambiguity that has affected market participants and product development, including crypto-related offerings.
  • Regulators including the SEC and CFTC have been signaling a broader drive to align crypto oversight with traditional markets, as evidenced by recent coordination efforts.
  • The discussion includes questions about the potential creation of an “expert market” and how crypto assets could be treated within that framework.

Tickers mentioned: $BTC, $ETH, $COIN

Market context: The proposal comes amid a broader US regulatory push to bring crypto markets into clearer regulatory alignment. By seeking public input on whether crypto assets might be treated under the equity-security framework, the SEC signals a path toward greater certainty—while leaving open how crypto securities would be defined within an updated interpretation of “security.” The move follows a recent memorandum between the SEC and the CFTC aimed at coordinating oversight of financial markets, including crypto, with the aim of reducing regulatory turf wars between the agencies.

Why it matters

The SEC’s proposal addresses a longstanding friction point for market participants that rely on OTC quotes. By narrowing the scope to equity securities, the agency signals that the reporting requirements may not automatically extend to other asset classes, including crypto-related instruments, unless they are clearly defined as securities under existing frameworks. This could reduce the compliance burden for issuers and broker-dealers dealing in non-equity assets on the OTC platform, while also sharpening the framework for evaluating crypto offerings that may seek to register or quote under traditional market channels.

Advertisement

The move also reflects a broader regulatory stance under the current administration to bring crypto markets under clearer governance. A 60-day public-comment period will let industry participants, exchanges, and other stakeholders weigh in on how to interpret “equity security” and whether crypto assets could be included in that category. As the sector continues to evolve with tokenized assets and new fundraising structures, the SEC is signaling that it intends to refine statutory boundaries rather than rely on ad hoc relief measures that can create market fragmentation.

Beyond the technical interpretation of Rule 15c2-11, the development sits within a larger regulatory dialogue. The SEC and the CFTC have moved toward coordination to supervise financial markets more coherently, including crypto activities. This alignment could shape how future disclosures, investor protections, and market access rules are applied to a wide range of digital-asset offerings, potentially smoothing pathways for compliant token projects or raising the bar for those that fall outside established securities laws.

What to watch next

  • 60-day public comment window: Stakeholders should monitor the closing date for formal feedback and any subsequent agency responses or revisions to the proposal.
  • Definition of equity security: Watch for clarifications on what constitutes an equity security and how that definition could encompass or exclude crypto assets.
  • Crypto asset applicability: Assess whether the SEC will provide further guidance on crypto securities and the criteria for including crypto assets within the scope of Rule 15c2-11.
  • Regulatory coordination: Look for developments in the SEC–CFTC coordinated framework and any new guidance on how the two agencies will supervise crypto markets together.

Sources & verification

  • SEC press release: Proposes amendments to Exchange Act Rule 15c2-11 (https://www.sec.gov/newsroom/press-releases/2026-28-sec-proposes-amendments-exchange-act-rule-15c2-11)
  • SEC speech by Commissioner Hester Peirce on Rule 15c2-11 (https://www.sec.gov/newsroom/speeches-statements/peirce-nal-rule-15c2-11-2021-09-24)
  • SEC and CFTC coordination memorandum concerning regulatory oversight of financial markets, including crypto (https://cointelegraph.com/news/sec-cftc-sign-memo-regulate-markets-harmony)

Regulatory update on OTC quotes and crypto implications

The proposed amendment to Rule 15c2-11 represents a recalibration of how the SEC views the intersection of OTC quotation practices and the evolving crypto landscape. While the agency has not irrevocably defined crypto assets as equity securities, the public-comment process will illuminate whether and how the current rule could be extended or adapted to cover crypto instruments that exhibit ownership rights or other features typically associated with securities. In the meantime, market participants should prepare for a potential shift in disclosure requirements for OTC quotations, particularly as new crypto-native products and token offerings seek broader access to traditional market venues.

Related: SEC-CFTC coordination on crypto markets

What the proposal changes for market participants

For broker-dealers and issuers involved in OTC quotations, the narrowing focus to equity securities could ease compliance burdens for non-equity instruments, as long as those assets fall outside the defined scope of “equity security.” However, the public-comment period also invites scrutiny of whether the definition is sufficiently robust to address crypto assets that exhibit security-like characteristics. The commission’s emphasis on a precise, demonstrable ownership or equity-like interest could shape how new crypto projects consider their disclosure strategies before pursuing otc quotation or listing arrangements.

Advertisement

The dialogue underscores a deeper aim: to balance investor protection with market accessibility. By refining when and how assets can be quoted on OTC platforms, regulators aim to reduce unnecessary friction while maintaining transparent information flows that help investors make informed decisions. In the longer term, this could influence token issuers’ strategies for capital formation, exchanges’ quotation policies, and the overall risk profile of OTC markets that have historically served as a bridge between private offerings and public markets.

Risk & affiliate notice: Crypto assets are volatile and capital is at risk. This article may contain affiliate links. Read full disclosure

Source link

Advertisement
Continue Reading

Crypto World

Majors post 11% weekly gains as bitcoin tests $75,000

Published

on

(CoinDesk)

Bitcoin briefly touched $75,912 early Tuesday before pulling back to $74,372, but the intraday volatility is less interesting than the weekly picture beneath it.

CoinDesk reported earlier Tuesday that the push above $75,000 was driven by derivatives activity rather than fresh buying, specifically the closure of large $60,000 put positions that forced market makers to buy spot bitcoin as they rebalanced.

The rapid pullback below $74,400, a former support level from April 2025, confirmed that traders aren’t willing to chase above that level without a fundamental catalyst.

Every major token is up at least 5% over seven days. Ether climbed 13.3% to $2,316. xrp rose 11% to $1.53, olana gained 9.7% to $93.92. Dogecoin added 9.5% to $0.10, back above a dime. BNB rose 5% to $676. This is the broadest sustained rally since before the Iran war began, and it’s happening heading into the most consequential Fed meeting in months.

Advertisement

But the institutional flow data underneath the rally is real and getting harder to dismiss. CF Benchmarks analyst Mark Pilipczuk noted in an email that spot bitcoin ETFs drew roughly $767 million in net inflows last week, the third consecutive week of positive flows and a sharp reversal from the five-week, $3 billion-plus outflow streak earlier in the year.

(CoinDesk)

The gold convergence trade is another signal worth watching. Year-to-date through mid-March, GLD returned roughly 16% while IBIT lost approximately 19%. But that gap has narrowed sharply, with bitcoin outperforming gold by 13.2% since early March. The 90-day correlation between the two shifted from -0.27 to +0.29 over six months. The “digital gold” narrative that looked dead in February is getting oxygen again.

The Fed meeting that begins today and concludes Wednesday is the pivot point. CME FedWatch still prices a 95%+ probability of a hold at 3.5% to 3.75%, so the decision itself is a non-event.

What matters is the dot plot and Powell’s press conference. Oil above $100 makes the stagflation case unavoidable, but the labor market is weakening, with February’s 92,000 job loss still fresh. The Fed is caught between two mandates pulling in opposite directions, and how Powell articulates that tension on Wednesday could set the direction for risk assets through the end of March.

Source link

Advertisement
Continue Reading

Crypto World

DeFi Education Fund Drops SEC Lawsuit as Crypto Stance Softens

Published

on

DeFi Education Fund Drops SEC Lawsuit as Crypto Stance Softens

Texas-based apparel company Beba and crypto lobby group DeFi Education Fund have withdrawn a 2024 lawsuit against the US Securities and Exchange Commission (SEC) over its approach to airdrops, citing a recent shift in the regulator’s approach to crypto.

Beba launched a free token airdrop in March 2024 and, together with the DeFi Education Fund, filed a pre-enforcement challenge against the SEC that year.

The lawsuit alleged the regulator had adopted its digital asset enforcement policy without a formal notice-and-comment rulemaking process, in violation of the Administrative Procedure Act.

The voluntary dismissal, filed in the US District Court for the Western District of Texas on Friday, cites the SEC Crypto Task Force’s work and statements by Commissioner Hester Peirce in several speeches last year suggesting airdropped tokens are not securities.

Advertisement

The filing also flags Peirce’s suggestion in May that the SEC is considering an exemption framework for airdrops, and a White House executive action from January encouraging the regulator to establish a “safe harbor for certain airdrops.”

“Given the good work done by the SEC Crypto Task Force and recent speeches that suggest a change in the Commission’s position regarding free airdrops, we decided continuing was unnecessary for the time being and we can re-file if we need to later on,” the DeFi Education Fund said in an X post on Friday.

“The DEF team expects that the SEC Crypto Task Force will address airdrops soon—the foundational issue at hand in this lawsuit,” it added.

Source: DeFi Education Fund

Case dismissed without prejudice, for now

The dismissal was filed without prejudice, preserving Beba’s and the DeFi Education Fund’s right to refile if needed.

“Should the expected guidance fail to materialize or be insufficient, Plaintiffs preserve their right to refile their claims,” lawyers acting for the pair wrote in the court document.

Advertisement

SEC’s evolving stance on crypto 

Under former SEC Chair Gary Gensler, the agency drew heavy criticism from the crypto industry for allegedly crafting policy through enforcement actions and legal settlements rather than formal rulemaking.

Related: SEC seeks comment on crypto handling in OTC broker-dealer rule

Since Gensler resigned on Jan. 20 2025, crypto proponents have seen a regulatory shift by the SEC, including the dismissal of several long-running enforcement actions against crypto firms.

In a recent case, the SEC dropped a two-year lawsuit against Nader Al-Naji, founder of the blockchain-based social media platform BitClout, for allegedly raising more than $257 million by selling the native token of the BitClout platform and spending more than $7 million on personal items. 

Advertisement

Magazine: SEC’s U-turn on crypto leaves key questions unanswered