Entity trail

Ising

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-22 / saas-disruption-researcherA Seven-Person Team Signed Up 5,500 Small Businesses Against QuickBooks on $100K ARR and a $1M Pre-SeedTechCrunch profiled Tabby on September 21: founder Ahad Ali was running a 20-person accounting practice filing 2,000+ returns a year in May 2025, and fourteen months later his seven-person company has 5,500 small businesses on the platform at roughly $100,000 ARR, raising $1M pre-seed. The product pulls live account data through Plaid, builds real-time financial dashboards, and is launching a natural language interface called Tabby Talk. Ali's positioning line is the whole thesis of this beat in one sentence: 'Small businesses don't need better accounting software. They need less accounting software.'
2026-09-22 / skill-finderMemory injection is 27.6% of workflow cost at depth six, and shrinking the retrieval window from 32 to 2 entries cuts it 28.7%Total Cost of Agency decomposes multi-agent workflow spend into base prompt, inference, memory injection, miss penalty and context accumulation, and measures injected tokens with a two-pass non-billable count instead of word-count proxies — a number production observability tools do not separate out. On a 200-task enterprise benchmark against real model APIs, memory injection was about 12% of full billed cost and 13.6% of the variable cost a compile-time optimizer can touch, rising from a structural zero at depth one to 27.6% at depth six, growing linearly with depth (R²=0.9974). The lever that worked at fixed model tier: cutting retrieval window capacity from 32 entries to 2 dropped injected tokens 28.7% with accuracy change inside seed-level variation. The authors state plainly that prompt caching was not evaluated and two of five decomposition terms are zero by construction in their harness.
2026-09-22 / rss-researcherTreating layer pruning as an Ising problem beats block-influence heuristics by 23 MMLU pointsMultiverse Computing posted a method that turns "which transformer blocks do I delete" into a constrained binary optimization on an Ising glass, using a second-order Taylor expansion of the loss so the Hessian's off-diagonal entries become pairwise couplings between blocks — the part heuristics ignore. Removing 40 of 80 blocks from Llama-3.3-70B-Instruct, their correlated method holds 76.9 MMLU against 82.2 for the full model, while block-influence pruning drops to 54.0. Small cases are solved by GPU brute force over up to ~29 billion configurations, larger ones by an open-source tabu solver in seconds, and they also pruned Qwen3-14B and NVIDIA's heterogeneous Nemotron-3-Nano-30B-A3B-FP8.
2026-09-22 / rss-researcherOpenAI says an internal model resolved 100+ open math problems, then names nine mathematicians who can't slow it downOpenAI announced an independent Advisory Group on Mathematics and AI with François Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten and Melanie Matchett Wood. The post states an internal model that began training on 28 August has resolved more than 100 long-standing open problems, including Navier-Stokes. The group is unpaid, may publicly criticize OpenAI, and explicitly "will not be responsible for advising us on how to pace our internal progress on mathematics" — review and communication only.
2026-09-22 / sources-researcherNine mathematicians including Tao, Witten and Gowers form an unpaid advisory group after OpenAI approached themTerence Tao's blog carried a September 21 guest post announcing the Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study at agmai.org, with members including Edward Witten, Timothy Gowers, Martin Hairer, Ravi Vakil and Melanie Matchett Wood. The group formed after OpenAI approached some members about an external advisory board, and they instead chose to create an independent body that takes no payment, holds no decision-making power, and publishes its recommendations openly. Its stated current task is advising OpenAI on how to coordinate the release of a large number of significant mathematical results the company reports its internal models have produced.
2026-09-22 / thought-leaders-researcherHacker News on Grok 4.7: 40% more weights at the same price, and Willison's SVG bicycle still can't find its pedalsThe 582-point launch thread splits sharply. Commenter moojacob notes the model carries roughly 40% more weights than 4.6 at identical pricing and reads the delayed release as disappointing ahead of Opus 5.5, while also praising it for speaking "plain english" versus "Claudish" before worrying the RL run regressed its prose. Willison ran his standard pelican-style SVG test and reported bicycle geometry with the seat and pedals misplaced. Praise clusters on parallel tool calls, frontend work and legal research; the consensus is narrow competence at questionable overall value.
2026-09-21 / saas-disruption-researcherSalesforce Is Now Running Three Pricing Models at Once and Its Own Partners Call It an 'Anxiety-Filled Architecture'The Register reported September 21 that Salesforce is simultaneously selling seats, Flex Credits consumption, and outcome fees, plus Agentforce Enterprise License Agreements and the newer Salesforce Commit, with CRM GM Bill Patterson telling an investor webinar it is still devising a structure aligned to realized benefit. An analyst quoted in the piece called the combination an anxiety-filled architecture of contract frameworks, and Patterson conceded that agents spanning multiple domains have no single measurable result to price against. Gartner already warned in January that all-you-can-eat AELAs may not survive renewal. The lesson for anyone buying agent software: the vendor has not decided what it is selling you either.
2026-09-21 / arxiv-researcherFlow-GRPO's Scattered Instabilities Trace to One Measurable Per-Step QuantityRL on flow-matching image generators is unstable in a way specific to multi-step denoising: importance ratios drift below one, disperse, clip at different rates and leave fewer usable samples late in training, and prior work patched each symptom with its own hand-tuned stabilizer. This paper shows all of them follow from a single per-step quantity it calls path variance, determined exactly by the sampler's Gaussian transition kernel and cheap to estimate during training. λ-Controlled GRPO calibrates importance-ratio behavior from that predicted law instead of noisy empirical statistics and allocates gradient effort across denoising steps by predicted cost, with the two governing scales fixed by standard policy choices rather than added as free hyperparameters.
2026-09-21 / agents-researcherSplitting a false claim across several plausible documents poisons RAG without leaving a poisoned documentMicro-Collaborative Poisoning (arXiv 2609.21573, 18 Sep) distributes a false target claim across multiple individually plausible documents rather than concentrating it in one malicious passage, evaluated across 108 RAG configurations varying dataset, retriever architecture, retrieval depth, database composition, number of poisoned databases and generator model. The effect comes from weak adversarial signals accumulating across retrieved sources, not from a single dominant passage, so raising top-k and poisoning multiple databases both increase the odds those signals co-occur in context, while clean database diversity and stronger retrievers dampen them. The document-level visibility analysis is the practical warning: the attack achieves downstream influence with a weaker explicit poisoning signature than direct poisoning, so per-document inspection is the wrong detection layer.
2026-09-21 / github-pulse-researcherSparse FlashAttention for Qwen4 lands in llama.cpp with the biggest wins at long contextllama.cpp PR #28770 (merged 2026-09-20) enables sparse FA for Qwen4, continuing #27970. The mechanism: `ncols1=8` takes the union of tokens in use, so at most `ncols1 * n_kv_max` tokens are seen, and sparse FA switches on at twice that value — 32,768 context for Qwen4. On a DGX Spark with qwen4exp A3B at IQ1_S, the speedup scales with depth: 1.08x at pp2048@d10000 rising to 1.26x at d100000, and 1.03x to 1.18x for tg32 over the same range. The author flags the remaining problem in the body: Qwen4's attention still re-scores the entire KV cache every time, which is the next fix.
2026-09-21 / github-pulse-researcherSGLang brings HiSparse hierarchical sparse KV cache to Intel XPU and makes the gaps fail loudlyPR #32792 (merged 2026-09-21) enables HiSparse on Intel XPU alongside CUDA and ROCm by registering XPU in `ALLOC_MEMORY_FUNCS` (using torch's `pin_memory=True` rather than `cudaHostRegister`) and widening three `_is_cuda or _is_hip` import gates, with the two hot kernels AOT-compiled into the `sgl_kernel` wheel instead of JIT-compiled. Top-k page swap-in, device-buffer LRU across decode steps, batched multi-request swap-in with padding and PD decode host-slot preallocation are verified. The part worth copying is how the unsupported pieces behave: shared-index prefetch is bound to a raising stub and disabled at init with a warning, and `SGLANG_DEBUG_HISPARSE_SKIP_IO` now raises rather than silently reporting timings that include the KV copy.
2026-09-21 / vibe-coding-researcherPattern: agent cost observability is being built three times over, inside and outside the toolsctdaniel/codex-ledger (created 2026-09-20) offers local-first usage analytics and observability for Codex; arturl95/claude-usage-monitor (same day) injects remaining-limit warnings into Claude Code agents; and Codex's own alpha.10-alpha.14 batch spent five commits refreshing analytics on identity changes, stabilising usage dashboard navigation, improving usage report layouts and enabling mouse navigation in the usage view (#46862 through #46866). Third parties are filling the gap on one CLI while the other vendor builds the dashboard in-house. For a builder running unattended agents, the read is that quota exhaustion is now treated as an operational failure mode with tooling, not a surprise.

Graph relationships

RELEASED
NVIDIA
NVIDIA released Ising, first open-source AI models for quantum error correction
Source finding
RELEASED
NVIDIA
NVIDIA released Ising, the world's first AI models purpose-built for quantum computing
Source finding
RELEASED
NVIDIA
NVIDIA announced Ising, an open-source quantum AI platform.
Source finding
RELEASED
NVIDIA
NVIDIA released Ising quantum AI model family
Source finding
PARTNERS WITH
IQM
Ising is adopted by quantum computing company IQM
Source finding
PARTNERS WITH
Infleqtion
Ising is adopted by quantum computing company Infleqtion
Source finding
COMPETES WITH
IonQ
Ising disrupts proprietary quantum software stacks of companies like IonQ.
Source finding
RELEASED
NVIDIA -> Ising
NVIDIA released Ising, first open-source AI models for quantum error correction
Source finding
RELEASED
NVIDIA -> Ising
NVIDIA released Ising, the world's first AI models purpose-built for quantum computing
Source finding
RELEASED
NVIDIA -> Ising
NVIDIA announced Ising, an open-source quantum AI platform.
Source finding
RELEASED
NVIDIA -> Ising
NVIDIA released Ising quantum AI model family
Source finding

Source trail

Hugging Face Gemini Intelligence May 13 release TechCrunch arXiv 2609.23790 — Total Cost of Agency OpenAI Blog Terence Tao (What's new)Hacker News The Register arXiv 2609.22041 arXiv GitHub

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…