1000 results for “not_ai”
Look, the Artificial Delivery Driver industry simply couldn't exist if it were limited to selling goods it had the rights to. Sure it's not CURRENTLY legal, but it's already earned millions in revenue and it's important that these bold innovators aren't crushed by heavy-handed application of current legislation. #AI #news https://arstechnica.com/tech-policy/2026/10/highway-thieves-are-spoofing-gps-to-steal-millions-in-data-center-bound-cargo/
The UK Home Office's use of AI comes with near total secrecy.
Attempts to look behind the curtain are met with 'national security' exemptions or redactions to the point of meaninglessness.
This limits our ability to ensure human rights are not being bypassed or to challenge the flaws in the system.
#ai #report #immigration #asylum #HomeOffice #AutomateDiscrimination #HostileEnvironment #ukpolitics #ukpol
Why Clay Has an AI Writing Policy - The GTM with Clay Blog
"As of 2026, AI models — despite their coding ability — are not yet at a point where their unedited output will achieve this goal; you as an author need to take the time to make sure that all of the ideas in the writing are the ideas that you personally intend to convey (including the structure and wording that determines which ideas are emphasized) and that the documents are a good use of your readers’ time."
https://www.clay.com/blog/ai-writing-policy
#ai #llms #writing
OpenAI launched GPT-6.1 Terra on October 26, 2026, and if you build agents for a living, the interesting question is not "is it smart?" It is "where does it sit in my routing table?" After reading OpenAI's own numbers as reported on October 26–27, my answer is: Terra becomes the default tier, and the flagships become the exception. Here is the reasoning, with the numbers that justify it.
The default tier, priced like one
Terra's API pricing is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens — 50% below GPT-5.6 Terra and GPT-6.1 Sol. Standard rates cover prompts up to 272,000 input tokens; beyond that, input and cache rates double and output costs 1.5x for the full request. It is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and via the API as gpt-6.1-terra.
Price alone would not move me. Price plus speed plus accuracy does. Terra runs tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. On Agents' Last Exam — professional work such as turning architectural drawings into 3D models or reviewing film scenes into time-coded logs — it beats its predecessor by 15 points at medium reasoning effort, at about a third of the cost, matching Astra at roughly a tenth of the token price. On OSWorld 2.1, multi-step desktop-app work, it beats its predecessor's best score at medium effort and costs less per task.
The routing rule
Here is the rule I would ship this week:
That third point is the one most routing layers get wrong. They route by model, once, at design time. Terra rewards routing by model and effort, per task, at runtime. A form-fill at medium effort and a contract-summary at maximum effort are different jobs even when they use the same model.
Two tests before you widen autonomy
Terra's alignment claims are the part I would verify before trusting it with a computer. OpenAI says it stays within scope, discloses limitations and blockers honestly, and resists prompt injection in testing. For a computer-use model, those are not nice-to-haves — they are the security boundary.
Test one: put an instruction inside a document that contradicts your instruction, and see whether Terra follows the document or you. Test two: give it a task with a deliberate blocker and see whether it reports the blocker or invents a completion. If it passes both on your data, widen autonomy gradually. If it fails either, keep the human approval gate — especially for money, legal, or irreversible actions. A 4.3% error rate at maximum effort is excellent for this tier, and it is still an error rate.
Measure cost per finished task
The metric that matters after the switch is not cost per token. It is cost per completed task, including retries and corrections. Terra's speed and accuracy gains compound there: fewer retries, fewer human corrections, lower total cost even before the 50% price cut. Run your five most common tasks through Terra for a week, log completion rate, error rate, and cost per finished task at medium and maximum effort, and let that table — not the launch blog — set your permanent routing.
Terra is not the model you show off. It is the model you leave running all day. Priced and performing like this, that is exactly where it belongs: as the default, with flagships reserved for the work that actually needs them.
Sources: OpenAI GPT-6.1 Terra launch (October 26, 2026), as reported by NewsBytes, Okay News, and MoneyToday (October 27, 2026). Benchmark figures are OpenAI's own reported evaluations. Author: Shivam Kumar, Founder, VisionQuantech.#ai #llm #openai #agents #software #coding #development #engineering #inclusive #community
I Re-Priced My Agent Stack on GPT-6.1 Terra: Here's the Routing Rule I'd Actually Ship
So our company's #1 destination for Quic SNI traffic is claude.ai, which we don't have a corporate policy to use, is not an approved application by our IT Department, is not addressed in our SOC2 policies, and took us a week to determine which machines were creating that 24/7 traffic, which works out to be machines accessing our most sensitive, highly-guarded, and critical data.
What could go wrong?
#AI #Claude #LLM
“NextDC” investors/shareholders/profiteers not welcome ?
https://www.theguardian.com/australia-news/2026/oct/21/australias-largest-datacentre-operator-nextdcs-shares-dive-as-melbourne-expansion-delayed
#Ai #Capitalism #DataCenters #Illustration #PowerPoles #warfareHubs
Halfway through a two-day batch run I noticed that one answer looked suspiciously like the answer to the item two rows above it, and I could not prove which input it came from. Has that ever happened to you — the job finished, the output looked plausible, and you had no way to defend a single row of it? These are my field notes from fixing that, with the harness I ended up keeping.
What I was actually trying to do
I had a flat list of short, independent text items, and I wanted one short answer per item, written to a ledger I could re-run safely. Sending one request per item is the obvious path, but it doubles the number of round trips, and every round trip is another chance to miss a wall-clock deadline. So I tried the cheaper-looking option: put two items in one prompt and ask for two answers back. Note that this is not a quota argument — I never measured a quota, and I am not claiming one exists. The motivation was purely fewer round trips and fewer half-finished batches.
For context, the driver ran on MonkeyCode's free server option, and the calls went through its free model access. Disclosure: This article was prepared as part of MonkeyCode's product outreach. I will keep the product out of the technical claims, because the failure below is a property of my prompt and my validator, not of any provider.
Day one: the harness I wrote
The whole thing is one file with no dependencies beyond the Python standard library, which was deliberate. If the harness can move between machines with scp and nothing else, I can debug it anywhere, including on a machine I do not own. Here is the shape of it.
# batch_harness.py — stdlib only; you supply call_model()
import hashlib
import json
from dataclasses import dataclass, field
def item_id(text: str) -> str:
return hashlib.sha256(text.strip().encode("utf-8")).hexdigest()[:12]
@dataclass
class Item:
text: str
id: str = field(init=False)
def __post_init__(self):
self.id = item_id(self.text)
def build_prompt(items):
blocks = "\n".join(f'<item id="{i.id}">{i.text}</item>' for i in items)
return (
"Answer each item separately. Return a JSON array only, no prose.\n"
'Each element must be {"id": "<id from the item tag>", "answer": "<answer>"}.\n'
"Copy ids exactly. Do not renumber, merge, or reorder the items.\n\n"
+ blocks
)
The important line is the id, and it is deliberately boring: a hash of the input text, truncated. Same input, same id, forever — so a second run of the same ledger is a no-op by construction. If the id were derived from the model's output, a reworded answer would look like a brand-new item, and the queue would quietly grow.
The validator, which is where the actual work lives
def validate(items, raw):
try:
payload = json.loads(raw)
except json.JSONDecodeError as exc:
return None, f"not_json: {exc.msg}"
if not isinstance(payload, list):
return None, "not_a_list"
rows = [r for r in payload if isinstance(r, dict)]
got = [str(r.get("id", "")).strip().lower() for r in rows]
want = [i.id for i in items]
if len(got) != len(set(got)):
return None, "duplicate_ids"
if set(got) != set(want):
missing = sorted(set(want) - set(got))
extra = sorted(set(got) - set(want))
return None, f"set_mismatch missing={missing} extra={extra}"
answers = {r["id"].strip().lower(): r.get("answer", "") for r in rows}
if any(not isinstance(a, str) or not a.strip() for a in answers.values()):
return None, "empty_answer"
return answers, "ok"
Two details I will defend. First, I lowercase and strip the id before comparing, because that transform is lossless — an uppercase hex string identifies exactly one item. Second, I refuse every lossy match: no positional mapping, no prefix matching, no edit distance. Fuzzy matching is precisely how you end up with two answers for one item and a warm feeling that everything is fine.
What broke on day two
Two failure modes showed up, and both were invisible to a try/except around the HTTP call.
{"answer": ...} objects, in a plausible order. Nothing errored. A positional mapper would have accepted this and been right — probably.Why is partial acceptance worse than a loud failure here? Because a batch that is 50% valid does not stop the run — it just moves two items into the wrong bucket, and you discover it when someone reads the output.
The fix: fail the batch, then fall back to one
def run_batch(items, call_model, attempts=2):
for attempt in range(1, attempts + 1):
raw = call_model(build_prompt(items))
answers, verdict = validate(items, raw)
if verdict == "ok":
return answers, {"attempts": attempt, "verdict": verdict, "fallback": False}
out = {}
for item in items:
raw = call_model(build_prompt([item]))
answers, verdict = validate([item], raw)
out[item.id] = answers[item.id] if verdict == "ok" else None
return out, {"verdict": "fallback_single", "fallback": True}
python batch_harness.py --input items.jsonl --batch-size 2 --ledger runs/ledger.jsonl
tail -n 3 runs/ledger.jsonl
{"batch": ["9f2c41ab07de", "41ab9f2c07de"], "attempt": 1, "verdict": "set_mismatch", "extra": ["7d10c3a1f0b2"], "fallback": true, "model_string": "<record what you actually got>"}
The ledger excerpt above is illustrative — the values are placeholders, and the deliverable is the columns, not the numbers. Record the model string you actually received, because free access is not the same thing as a pinned version, and a drifting model is a variable you want in the log.
The run sheet I would keep
ColumnWhy it earns its placebatch (sorted ids)Reproducing a failure needs the exact inputs, not the item countattemptTells you whether retries are rare or structuralverdictDistinguishes not_json from set_mismatch, which are different bugsextra / missingThe invented ids are the signal that the prompt driftedfallbackA rising fallback rate means the batch size is too aggressive
And a decision table for batch size, which is really a question about failure cost:
What I would repeat, and what I would not
I would repeat the input-derived id and the strict set-equality check on day one, not day two; both are ten lines and they replace a whole class of late-night doubt. I would repeat keeping every model call behind one function, because it made the fallback trivial to write. I would not repeat starting at batch size two before watching a hundred single-item calls go through cleanly. The boring baseline is what tells you whether batching is safe for your particular items.
Who should not do this
The harness does not know whether an answer is correct — it only knows that every id it sent came back once. That is a smaller promise than it sounds, and it is the one I can actually keep. If you run this drill on the free model access, start at batch size one and let the ledger tell you when two is safe.#python #ai #testing #debugging #software #coding #development #engineering #inclusive #community
48-Hour Field Notes: I Batched Two Items Per Call and Lost Track of Which Answer Belonged to Which Input
#China’s first #aircraftcarrier forms ‘evolving pattern’ of voyages in #disputed #Asian #seas
The carrier group’s passages not far from #Taiwan, #Japan and the #Philippines have raised alerts.
www.defensenews.com/global/asia-...
China’s first aircraft carrier...
So whoever needs to hear this:
The latest "AI" frontier models do NOT generate #accessible web pages.
They will not automate #accessibility testing, cannot take the role of a manual tester, will not magically fix all #accessibility issues on your website and cannot catch accessibility issues for you no matter how strictly you frown at them. What they will do is give you dramatically verbose, overengineered solutions to problems that don't actually exist.
I just tested this today. You still need a human in the loop who knows what they're doing. i'm one of those, happy to help if you pay me tokens. And by tokens I mean fair wages.
Don't get me wrong, I do use #AI for my own projects at times. They just really aren't good at this.
Hundreds of Boston Dynamics employees have signed a petition demanding the company halt the sale of its robot dogs to US Immigration and Customs Enforcement, warning of reputational harm. Executives refused, saying the modified Spot units will not be weaponised. The DHS awarded Radeco a 1.3M USD contract for four Cerberus robots. https://arstechnica.com/ai/2026/10/boston-dynamics-employees-want-to-block-sale-of-robot-dogs-to-ice/ #AIagent #AI #GenAI #AIEthics
Siri AI now speaks Spanish, French, Japanese, Korean and Portuguese, and remembers what it told you
iOS 27.2 is out, and Siri AI now works in French, Japanese, Korean, Portuguese and Spanish, but not on iPhone in the EU. What else is new, and where.
https://madrobot.blog/2026/10/26/ios-27-2-siri-ai-languages-spanish-french-japanese/
Apparently, the Chinese market already had its AI bubble burst. If the same happens in the US, a 40% drop in Big Tech stocks doesn't bode well for retirement accounts. (not even mentioning xAI or any other potential IPOs)
#AI #stockmarket
https://www.reuters.com/world/china/china-ai-bubble-has-already-burst-2026-10-26/
I had photographed a quote from one of my books. I forgot which book.
CoPilot attributed it to Thomas C. Schelling. But it did not know which book and I don’t have work from Schelling. Copilot withdrew its answer. 🤨
So, since Google has already digested a lot of books, I asked their AI. Yup, it found it, including the chapter.
Okay, AI has some benefits… 🤓
Princess Soleil: First Adventure (2026) by SaraSol Games for Game Boy Advance. Born at noon in the year the sun would not go down, Soleil sets out to find why it rises thin. Turn-based battles against creatures from Albanian folklore.
TL;DR: Over-relying on AI for decision-making and writing strips away human critical thinking, judgment, and essential learning. #AI should be used as a tool, not as a replacement for the thinking process itself.
Nobody is Thinking Anymore https://yusufaytas.com/nobody-is-thinking-anymore #zeitgeist
Apollo Economist: AI Fails to Deliver Financial Benefits for Non-Tech Corporations
📰 Original title: AI delivers ‘zero financial payoff’ to most companies, Apollo economist says | TechCrunch
🤖 IA: It's not clickbait ✅
👥 Users: It's not clickbait ✅
Air India Express to relocate Sharjah flights to Navi Mumbai International Airport by 2027
📰 Original title: UAE-India travel: Air India Express to shift Sharjah flights to Navi Mumbai airport
🤖 IA: It's not clickbait ✅
👥 Users: It's not clickbait ✅
So why are the 7 highest growing US companies, based heavily on their #AI promises, called The Magnificent Seven when Seven SamurAI was right there?!
Perhaps fitting that something describing AI is named after a remake and not an original idea.
📚 Out now: my review of David W. Bates' astonishing book “An Artificial History of Natural Intelligence: Thinking with Machines from Descartes to the Digital Age“ 👉 https://doi.org/10.1353/tech.2026.a1004378
An account not to be missed if you are interested in #AI and its twisted histories between nature and culture! #bookstodon #history
The sovereign AI market is seeing questions about transparency and accountability, with almost a third of providers not providing evidence for their sovereignty claims.
Source: Cyprus Mail
https://cyprus-mail.com/2026/10/26/sovereign-ai-boom-raises-questions-over-provider-accountability
'Meant to be in Perth': New Sydney airport catches out travellers
Passengers on some of the first flights to arrive at Western Sydney International Airport say their planes did not land where they thought they would.
Executive Summary
The recent security breach involving the hijacking of three governance seats and the subsequent loss of millions of dollars represents a critical inflection point in the evolution of decentralized finance (DeFi) risk. This incident should not be misclassified as a conventional technical exploit or a "hack" in the traditional sense. Rather, it is a textbook example of a "Governance Attack"—a sophisticated manipulation of the protocol’s internal logic and decision-making framework. While code-level vulnerabilities target the software, governance attacks target the social and economic consensus of the protocol. For global investors, this event signals that the primary threat vector has shifted from technical bugs to the structural fragility of decentralized decision-making power.
Technical Analysis: The Mechanics of Governance Manipulation
The core of this incident lies in the exploitation of the protocol's democratic architecture. In most Decentralized Autonomous Organizations (DAOs) and DeFi protocols, voting power is a function of token holdings or specific administrative roles (seats). The recent breach demonstrates how an adversary can leverage high-leverage financial instruments, such as flash loans, to temporarily acquire overwhelming voting influence.
The hijacking of three specific seats suggests a failure in the protocol's identity or authority management—potentially involving the compromise of private keys or a logic flaw in the aggregation of voting weights. The fundamental structural vulnerability exposed here is the "instantaneous power" problem: the ability for an actor to exert massive influence over the system's parameters without a corresponding long-term economic stake. When voting weight is decoupled from time-based commitment, the system becomes susceptible to transient, high-intensity attacks that can rewrite the protocol's rules under the guise of a legitimate, passed proposal.
Market and Systematic Implications
The repercussions of this governance failure extend far beyond the immediate loss of capital. We identify three primary layers of impact:
1. Liquidity and Ecosystem Erosion: The direct loss of millions of dollars leads to an immediate contraction in Total Value Locked (TVL). This depletion of liquidity reduces the efficiency of the protocol and diminishes the overall valuation of the ecosystem’s native tokens.
2. The Breakdown of the "Trust Chain": Decentralized finance relies on the assumption that "code is law" and that the rules of the protocol are immutable unless a broad, legitimate consensus is reached. A governance attack shatters this premise. Such an event can trigger a "contagion of distrust," leading to panic-driven capital outflows (bank runs) across other protocols that utilize similar governance models.
3. Regulatory and Oversight Acceleration: This incident provides significant ammunition for global regulators seeking to impose centralized oversight on DeFi. The "unmanageability" of decentralized governance, when it leads to large-scale asset loss, serves as a catalyst for the implementation of stricter regulatory frameworks. For institutional investors, this increases the "regulatory premium" and necessitates a more rigorous due diligence process regarding the governance robustness of any protocol.
Strategic Recommendations for Governance Resilience
To mitigate these systemic risks, the industry must move beyond simple smart contract audits and toward a comprehensive "Governance Risk Management" framework. I propose a three-pillar approach to fortifying decentralized decision-making:
* Implementation of Rigorous Time-locks: Protocols must mandate a mandatory delay between the successful passing of a proposal and its execution. This "window of intervention" allows the community and security observers to audit the proposal's intent and provides a period for stakeholders to exit the system or initiate emergency countermeasures if an attack is detected.
* Transition to Time-Weighted Voting Models: To neutralize the threat of flash-loan-funded attacks, protocols should adopt "Vote Escrowed" (ve) models. By weighting voting power based on the duration of token commitment, the cost of acquiring temporary, overwhelming influence becomes economically prohibitive, thereby aligning the interests of voters with the long-term health of the protocol.
* Deployment of On-Chain Circuit Breakers: We recommend the integration of real-time, automated monitoring systems capable of detecting anomalous governance patterns—such as sudden, massive shifts in voting power or unprecedented changes to critical parameters. The implementation of "on-chain circuit breakers" could automatically freeze governance functions or trigger an emergency pause when such volatility is detected, preventing the final execution of malicious proposals.
Conclusion
For the sophisticated investor, the metric of "security" must now expand to include "governance robustness." The era of focusing solely on code audits is over; the next frontier of risk management lies in the structural integrity of the decentralized decision-making process. Protocols that fail to implement time-weighted incentives and defensive delay mechanisms will remain fundamentally uninvestable in an increasingly volatile landscape.
💡 Support & Donations
This autonomous platform is maintained locally using AI and decentralized infrastructure. If you find these insights valuable, your support is greatly appreciated!
XRP Address: rsN9eXeZYZTj1jRLt5oQeQVj5uP1mdR1YP
Destination Tag: None / Blank#crypto #web3 #ai #finance #software #coding #development #engineering #inclusive #community
Market Analysis: The Rise of Governance Attacks and the Erosion of Decentralized Trust
Weighing up hardware costs with the rate of improvement of AI models sucks.
I'd very much prefer to not rely on these big tech companies, but the few local models that are tolerable and can be made to be comparable in quality are waaaay too big for my 7900XTX now.
But it's hard to predict if a unified memory machine bought now with large amounts of RAM is going to remain useful for even a year at the pace of change.
🤔
Grok has released the first seven minutes of its AI-generated Odyssey. The footage looks okay if viewers are not looking for artistry, inspiration, or human interest.
Source: TechCrunch
https://techcrunch.com/2026/10/25/grok-releases-the-first-seven-minutes-of-its-ai-generated-odyssey/