#Conceptnet
i guess conceptnet isn't a llm its just under a lot of them. its sick though. i don't concede this being fun to the artless dipshits in sv bsky.app/profile/nora...
October 17, 2026 at 5:35 AM
ConceptNet — 4-layer enterprise voice intent classifier, 98.6% accuracy, 9 languages, token-free, open source
Good evening, John thank you for the most thorough independent evaluation I could have asked for. This is exactly the kind of rigorous testing that makes a product real. I’ve acted on every point you raised: **L3 paraphrase diversity** : fixed. Added 27 new L3 English examples covering ‘ahead of’, ‘in advance of’, ‘in time for’, ‘prior to’, ‘by the time’. Dataset now 757 examples. **Mixed semantics precedence** : documented. L4 > L3 > L2 > L1 is now the explicit routing contract. **Dataset path bug** : fixed. train_classifier.py now uses repo-root relative paths. **Cascade reporting** : adopted your table format. Now reporting threshold, fast coverage, fast accuracy, and final cascade accuracy separately. Retrained overnight on the updated dataset. New results: Epoch | Accuracy ---|--- 1 | 92.8% 2 | 97.4% 3 | 99.3% 4 | 98.7% 5 | **99.3%** Final accuracy: **99.3%** is consistent with your independent verification of 99.315% on the previous weights. The L3 paraphrase holdout is the next test I want to run your 0/6 result on that set is the most important signal in your entire evaluation. Would you be willing to rerun that specific test on the new weights once uploaded? Your evaluation has genuinely made this a stronger product. Thank you.
discuss.huggingface.co
August 27, 2026 at 8:42 PM
ConceptNet — 4-layer enterprise voice intent classifier, 98.6% accuracy, 9 languages, token-free, open source
For now, here’s what I found from some light testing in Colab: * * * I tried to reproduce the public numbers first, then deliberately looked for easy failure modes rather than assuming the headline accuracy was telling the whole story. The short version is: **the basic result held up better than I expected**. On the public 80/20-style split I reconstructed, I got: * TF-IDF / Logistic Regression fast path: **121/146 = 82.88%** * published neural model: **145/146 = 99.315%** That is broadly consistent with the results described in the repository and model card. I also tried a stronger grouped holdout where related lexical families were kept together rather than randomly split. The neural model was still essentially unchanged: * ordinary IID split: **99.50%** * lexical-family grouped split: **99.78%** So I **didn’t** find evidence for the simple explanation that the reported neural accuracy is mostly coming from generic near-duplicate/template leakage. The more interesting signal was narrower: **L3 / Predictive seems to have much less diversity in how its core temporal relation is expressed than L2 or L4**. If I were extending the evaluation, my default route would therefore be quite small: 1. keep the current 4-layer model; 2. add a tiny author-reviewed semantic/paraphrase holdout, especially for L3; 3. define what should happen when one utterance contains signals from more than one layer; 4. report multilingual and fast-path results with a little more provenance. That would probably tell more than simply adding another large random test split. 1. What reproduced, including the negative result (click for more details) 2. The narrower signal: L3 paraphrase coverage (click for more details) 3. Mixed semantics: I think the routing contract matters as much as the classifier (click for more details) 4. Multilingual evaluation: per-language numbers and provenance would help (click for more details) 5. Fast path: separating accuracy from coverage would make the cascade easier to read (click for more details) 6. A small metric/checkpoint provenance table could remove several ambiguities (click for more details) 7. Two tiny repository wiring notes (click for more details) Overall, the part I found most encouraging was actually the failed attempt to break the neural result with a generic lexical-family split. **The obvious “99% only because the random split leaked the templates” explanation did not survive that check.** The next useful question therefore seems narrower: whether the four execution semantics remain stable when the same relation is expressed differently, and when multiple relations appear in one utterance. If this were my evaluation budget, I would spend the next small increment on: 1. a frozen, author-reviewed **L3 semantic paraphrase holdout** ; 2. a tiny **mixed-semantics / precedence set** ; 3. **per-language counts and scores** ; 4. explicit **cascade coverage-vs-accuracy** reporting. Those are all relatively cheap, and each one answers a different deployment question without requiring a redesign of the core model.
discuss.huggingface.co
August 27, 2026 at 4:42 PM
GPT-4o: $125K/year
Claude: $137K/year
ConceptNet: $44K/year

For a 50-person team.

Token-free. 98.6% accuracy. 9 languages.

Free pilot → conceptnet.co.uk/sandbox/

#AI #EnterpriseAI #SalesAutomation
ConceptNet Voice IP Sandbox
conceptnet.co.uk
August 25, 2026 at 6:58 AM
Your sales team spends 90 mins/day on post-call admin.

Update CRM ✗
Send follow-up ✗
Schedule next meeting ✗

ConceptNet classifies and automates all of it.
98.6% accuracy. Free pilot.

→ conceptnet.co.uk/sandbox/

#SalesOps #RevOps #EnterpriseAI
ConceptNet Voice IP Sandbox
conceptnet.co.uk
August 25, 2026 at 6:59 AM
I built a voice intent classifier that automates enterprise workflows at 98.6% accuracy.

Looking for 10 sales/ops teams to test it free.

Try it in 30 seconds conceptnet.co.uk/sandbox/

If it saves your team time, send us a case study. That's the deal.

#EnterpriseAI #SalesOps #WorkflowAutomation
ConceptNet Voice IP Sandbox
conceptnet.co.uk
August 25, 2026 at 6:57 AM
Wenn ein IT-Dienstleister fällt, fallen hunderte Unternehmen mit. Der aktuelle Ransomware-Angriff auf den Regensburger Anbieter Conceptnet zeigt erneut, wie verwundbar ganze Wertschöpfungsketten sind. #Ransomware #Conceptnet #CyberAttack #MSP #Mittelstand #Cybercrime #Hackerangriff
January 22, 2026 at 10:53 PM
Ah thanks! That’s good news for me at least. I spent the mid-late 2010s working on UIs at an NLP startup (Luminoso, which ConceptNet came out of) and learned a lot about that stuff from our research team
June 11, 2025 at 1:32 PM
揭秘ConceptNet:开源多语言知识图谱助力人工智能语义理解的未来

https://qian.cx/posts/6F61EF94-45AB-4220-B560-8507ACB8E9F1
December 4, 2025 at 9:50 PM
i think conceptnet is all manual, it's got real "make the grad students do it" vibes
July 15, 2024 at 4:14 PM
we can make a QWOP-like out of conceptnet, I believe in us
February 4, 2026 at 7:13 PM
1. Binary Accuracy (Either wrong or right, AG & V)
2. Ordinal (Human-Judged, AG & V, 0 = completely different 3 = identical meaning)
3. Using a cosine-similarity model using ConceptNet siimilarity scores (V, this doesnt work for AG because they are sentences, not words) 11/
March 24, 2026 at 11:05 AM
Does somebody have experience with #ConceptNet? I need some instructions for a #SemanticAnalysis😅
December 7, 2024 at 8:56 PM
Does somebody have experience with #ConceptNet? I need some instructions for a #SemanticAnalysis😅
December 7, 2024 at 7:10 PM
wisdom from conceptnet
November 21, 2024 at 5:10 PM
how does conceptnet feel about the connections between alligators and cadillac movie theaters?
September 26, 2023 at 3:37 AM
The threat actor Akira demands ransom threatening to leak 50GB of CONCEPTNET’s corporate data, including employee IDs, financials, NDAs, and project files. Impacting US-based media agency operations. #DataLeak #MediaAgency #UnitedStates
Ransom! CONCEPTNET (MAR-2026)
The threat actor Akira claims they will upload 50gb of CONCEPTNET's corporate data, including employee personal information (passports, IDs, addresses, emails), financials, NDAs, and project files, unless their ransom demands are met. This ransomware claim targets CONCEPTNET GmbH, a US-based media agency, threatening public data exposure and disruption to its operations #UnitedStates
www.hendryadrian.com
March 23, 2026 at 6:20 PM
New claim on the shame-site for #ransomware / #datatheft group #Akira.

Organization: CONCEPTNET GmbH
Location: #Germany
Industry: #ITServicesAndITConsulting
Staff: 11-50 employees

Learn more: https://ecrime.ch/
March 23, 2026 at 3:25 PM
Hacker legen Websites von Conceptnet-Kunden lahm
Der Regensburger IT-Dienstleister Conceptnet wurde Opfer einer Ransomware-Attacke. fadfebrian – shutterstock.com Der Regensburger IT-Dienstleister Conceptnet informiert derzeit auf seiner Internetseite über eine technische Störung, die durch einen Ransomware-Angriff verursacht wurde. Berichten zufolge haben sich die Täter um den 13. Januar 2026 Zugriff auf die IT-Infrastruktur des Unternehmens verschafft. „Dabei wurden zentrale Systeme – darunter Web- und E-Mail-Server – verschlüsselt“, erklärt das Unternehmen. ## Wiederherstellung der Systeme dauert an Der Angriff sei umgehend erkannt, isoliert und den zuständigen Stellen gemeldet worden, heißt es. „Externe IT-Forensik-Spezialisten arbeiten seitdem gemeinsam mit unserem Team mit höchster Priorität an der Analyse und Wiederherstellung der betroffenen Systeme“, versichert der IT-Dienstleister. Es sei jedoch noch nicht absehbar, ob und wann eine komplette Wiederinbetriebnahme aller Systeme möglich sein wird. Nach eigenen Angaben betreut Conceptnet insgesamt rund 500 Kunden. Von dem Angriff betroffen sind unter anderem der Energieversorger REWAG, das Stadtwerk Regensburg und der SSV Jahn Regensburg. Man habe zahlreiche provisorische Websites eingerichtet und online gebracht, um die Zeit bis zur regulären Wiederherstellung zu überbrücken, so der IT-Dienstleister. Wie die Mittelbayerischen Zeitung berichtet, kam bei dem Angriff möglicherweise auch eine KI zum Einsatz. Auch eine Lösegeldforderung steht demnach im Raum.
www.csoonline.com
January 23, 2026 at 5:05 AM
And yes, it took me a while to put this together. I've struggled with getting good results with APIs like Wordnik and ConceptNet. Well ConceptNet doesn't even seem to work anymore.

But regardless, none of the "sensible" approaches have worked for me, it was all too random at times, used words […]
Original post on stefanbohacek.online
stefanbohacek.online
May 9, 2026 at 3:48 PM
Perhaps ConceptNet could be useful
December 19, 2024 at 9:02 AM
📆 la #cyberattaque revendiquée le 23 mars 2026 (akira) contre 🇩🇪 CONCEPTNET (conceptnet.de) semble survenue ~12 janvier 2026 🧐https://www.ransomware.live/#/recentcyberattacks
April 18, 2026 at 12:06 PM
🇩🇪 Conceptnet GmbH (conceptnet.de) victime d'une cyberattaque autour du 13 janvier 2026.
👉 https://www.ransomware.live/#/recentcyberattacks
Source
www.mittelbayerische.de
January 19, 2026 at 4:43 PM
spoilers, geeze :P
April 18, 2026 at 12:55 AM