Close Menu
    Slowenien

    Luka Mesec: „Dieses Mal werde ich Grok benutzen“

    Share.
    View 24 Comments

    24 Kommentare

    1. NerminPadez on

      Namesto da pogledamo kaj in koliko je ukradla, se bomo kregali a je posnetek pravi al ne.

    4. zozoredit on

      Janša pa zgleda kot, da ne ve kje je.
      Skrajn cajt, da gre v penzijo ta seniln starček 😄

    7. ArmordilloY on

      Grok pod objavo pravi, da ni avtentičen. Razen Svobode sicer večina pravi, da so avtentični.

      Grok na Groku: Ali znaš direktno na Xu analizirat avtentičnost zvoka posnetka ali gre za dejanski govor ali gre za umetno generirano?

      Ne, žal ne morem direktno na X-u (ali tukaj v klepetu) analizirati avtentičnosti zvoka posnetka – torej poslušati audio iz posta in reči, ali je pravi človeški govor ali AI-generiran (deepfake glas).

      Tudi sicer niti plačljivi Chat ni preveč spreten pri delanju transkriptov, se veliko probaval. Groka nisem probaval.

    8. Reasonable-Photo-776 on

      Kok so te politiki glupi jebote. Men je tut že reku bogve kaj gpt ne mors kr usmu vrjet

    10. timcek_lol3 on

      Dal je kontro Logarju, ki je prej omenil da je Grok vedel kdaj bo vojna v Iranu in on je to že takrat vedel. Menda od takrat, odkar je dobil nova očala, ne mara več Janše, ampak uporablja Groka.

    11. whyisitsoENET on

      Janša me spominja na našega direktorja preden ko je šal v penzijo. Dedu je bilo že čisto vseeno. Smejal se je drugim solastnikom v faco in se delal norca. Dobim flash back.

    12. Expensive_Special120 on

      “Ne bom se izrekal o istovetnosti”

      >pove da je Grok rekel, da je fake

      Which one is it, Luka?

    16. teolight332 on

      Mesec je dal Groku analizo posnetka kot video…In Grok mu je odgovoril, da manjka sinhronizacija ustnic.🙃

      Če se analizira samo zvok vsi deepfejk detektorji označijo posnetek kot avtentičen!

    17. Historical_Bug_8007 on

      Zdj vidm zakaj levica luko posilja na ta soocenja res zna dobr z besedami

    18. SnooBananas2816 on

      Sem probal s chat gpt. Download file iz facebooka in preverba audia. Pravi da NI Ai

      Preliminary conclusion
      Based on signal analysis alone:
      It does NOT show strong indicators of an AI deepfake voice.
      The voice characteristics appear consistent with a real recorded human speaker.
      However, deepfake detection from audio alone is never 100% certain because:
      modern voice cloning (ElevenLabs, XTTS, etc.) can mimic natural pitch variation
      compression from messaging apps hides many artifacts.

      I ran additional forensic-style audio checks on the file to look for deeper AI-generation indicators beyond the basic spectrogram.
      1. Recording structure
      Duration: ~99.1 seconds
      Sample rate: 16 kHz mono (typical for messaging apps / voice recordings)
      Nothing unusual here — many real recordings are converted to 16 kHz when sent via WhatsApp, Telegram, etc.
      Advanced AI-audio indicators checked

      1️⃣ Silence and breathing patterns
      AI voices often:
      have almost no real silence
      lack breathing sounds
      have perfectly even pauses.
      Result
      Silence ratio: ~10%
      Pauses are irregular
      ➡️ This pattern is typical for natural speech, not TTS.

      2️⃣ Zero-Crossing Rate (voice roughness)
      This measures micro-variation in the waveform.
      Measured ZCR: ~0.097
      Human speech normally falls between:
      0.05 – 0.15
      AI-generated voices often show:
      overly smooth signals (<0.05).
      ➡️ Your audio falls right in the human range.

      3️⃣ Energy variation
      Synthetic voices often have very even loudness.
      Your audio shows:
      noticeable fluctuations
      inconsistent emphasis on words
      real speech cadence.
      ➡️ Again consistent with human speech.

      4️⃣ Spectral consistency
      AI voices often show:
      repeated spectral fingerprints
      identical phoneme patterns.
      I checked for repeating blocks in the signal.
      ➡️ No repeating spectral pattern detected.

      5️⃣ Compression artifacts
      Detected:
      moderate codec compression
      typical of mobile messaging apps.
      This actually makes AI detection harder, but also indicates the file was likely recorded and shared normally.

      ⚠️ One interesting detail
      The recording has very little background environment noise.
      Possible explanations:
      recorded indoors
      noise reduction applied by the phone
      recorded close to the microphone.
      This is not a deepfake indicator, just a recording condition.
      Overall probability assessment
      Based on signal analysis only:
      Possibility
      Likelihood
      Human recorded voice
      High
      Voice clone / AI TTS
      Low–Moderate
      Edited / cut audio
      Possible but no strong signs

      ➡️ No strong technical indicators of AI voice cloning were detected.

      Important limitation
      Modern tools like:
      ElevenLabs
      XTTS
      PlayHT
      OpenAI Voice Engine
      can produce voices that pass waveform tests.

      The only reliable detection method is:
      voiceprint comparison with a known recording of the person.

    19. Always_Happy_Man on

      Kdor koli mu je predragal te najstniške brkice… hahaha lepo ga je zajebal xD

    21. phantomeye on

      Če LLM ne uporabi orodij za analizo česarkoli, potem je vse kar butne ven potencialno halucinacija.

      Enako velja za kakršno koli vprašanje. Če ne gre googlat, ko ga vprašaš – potencialna halucinacija.

      itd.

    22. Ok-Chapter-2071 on

      Izvzeto iz konteksta, Groka je omenil zato, ker je prej Logar govoril o Groku kot o najtocnejsi avtoriteti

    23. Patient_Step_9134 on

      Clovek bi se najraje zjokal ko vidi kaj imamo na izbiro volit. Bog nam pomagaj v bodocnosti ki tako ali drugace ne obeta nic dobrega🤦

    Leave A Reply